If you’d like more of my musings, subscribe below:

Subscribe now

Spotify Discover fed me a new DJ this week. Tiny little artist with fewer than 100K listens. Perfect kind of lounge house vibes that I love.

I was smitten. Hell yeah, a diamond in the rough.

So, of course, being the music nerd that I am, I start Googling to get to the bottom of who this no- name creator is. They’ve got a dozen tracks on Spotify, yet they’re a ghost online. No Instagram, no Twitter, nothing. The most I can find is their songs cross-listed on Apple Music and YouTube as well. Weird.

Equally peculiar, they’ve got like four albums in their catalogue, but they all seem to have been released in May of last year…

I dig a little deeper.

The single shred of evidence I find is a deleted, cross-posted Reddit link:

Oh no. No, no, no, no.

Here I am bopping to my new favorite DJ for six hours, and suddenly I look up to find myself face to face with this guy:

And honestly, I felt alittle Will Smith:

I was slightly shaken.

I’m usually a proponent of what AI tooling is set to unlock. I previously wrote about The Arc of Accessibility, and how so many great consumer products got massive by empowering individuals to create in new and previously unseen ways. With music specifically, Ableton allowed a teen to create Grammy-caliber tracks in his bedroom and the internet enabled him to distribute that music infinitely. An incredible unlock for all of our earballs.

So, now I’m sitting here asking myself, “Well, if it was an individual leveraging AI tools to make more, better, and different music than they ever could have otherwise, that’s pretty sweet. More creativity in the world, and more tracks that I love!”

But if it was just end-to-end AI production generated en masse…

Well, shit. Now, I’m left wondering if I’m just another little piggy at the AI slop trough:

Because what does this mean for the future of music? For the future of art?

If these songs still evoke emotion in me, then what? Is it still slop?

I honestly can’t tell if this curve is correct or if it should be precisely the inverse. Perhaps I’m a mouth breather eager to be supplicated by the cultural opiate of infinitely-produced entertainment. Or maybe the slop mania really is the implosion of culture as we know it.

Granted, these tracks I’m discussing are a very specific Wynn flavor of repetitive kinda background music that would be fairly easy to replicate with AI, but these tools are advancing exponentially and will not be constrained to more formulaic genres for long. We will see a top charting track, completely AI generated, within the next twelve months. Mark my confused words. This will precipitate all sorts of questions that we’ll collectively be grappling with over the coming months and years.

For example, how does this evolve the position of an artist? Does this amplify the value of curation? In a world where artists are capable of making any style of music, freed of genre and production limits, does that compound the extent to which we’re really just valuing an artist for their taste more than their craft?

When I go to see a DJ at Concourse Project here in Austin, even the headliners probably only play 70% of their own music in an hour and a half long set. Half of the value is the in-person experience, but there’s also an element of enjoying their taste and the upcoming artists that they surface. I appreciate when Disclosure includes a tune from some DJ with 10,000 listens; it reminds me of just how relentless the pros are in pursuing new sounds.

Maybe this is overintellectualizing. The biggest artists in the world already become more brand than human today. Drake and Taylor Swift have an entire stable of songwriters, ghost producers, support staff, etc. to help perpetuate the industrial music machines they represent.

Plus, this dynamic holds across most creative arenas.

You go to a Michelin-starred restaurant to experience the ideas and menus of its famous head chef, but it’s his or her understudies that are plating your overpriced pork shoulder. A famously controversial example of this in the art world is one Jeff Koons:

“After becoming an established artist, Jeff Koons went on to establish a busy workshop space in the Chelsea district of New York. Here he employs a team of over 100 highly skilled assistants who make his art for him… [He] often creates the designs for his artworks by using computer technology, building how he wants the work to look before handing these digital prototypes over to his studio assistants, or other specialists.” - Rosie Lesso, How Does Jeff Koons Make His Art

In our Brave New World, how much does it matter that those starving NYU grad students are replaced with AI? You were paying $5M for the brand on a piece that Jeff’s hand may never have physically touched anyway. AI is enabling a deluge of slop, yes, but perhaps the more things change the more they’re really staying the same. We will latch onto artists and personalities for their taste, regardless of whether the creative’s team is underpaid assistants or autonomous Clawdbots. The biggest creators in the world aren’t businessmen, they’re a business, man.

That said, I’m still wrestling with the synthetic nature of it, I suppose. At least Koons has a human in the loop somewhere to execute his vision. Creativity and taste were promised to be one of the sacred bastions that makes us human, and we’re seeing this facade begin to crack:

It’s creative output industrialized.

What comes next is going to get weird. I’m not entirely sure how this plays out from a consumption/production perspective (how many AI DJs will I inevitably add to my rotation?), but I do have initial thoughts for what remains evergreen in this cultural avalanche: exclusivity and belonging.

Abundance in anything (art, music, etc.) will continue to drive value to that which is scarce. This cultural shift may create a space for more bespoke output– proof of humanity or tracks inclusive of the little bumps and imperfections that come with unprocessed creation. Perhaps, more live albums. But I think that is less likely to be the major change in music simply because widespread distribution is so key for an artist’s success. For example, it’s great that Wu Tang released a single print $2M Gold Album, except that no one ever listened to it besides like five crypto bros and Martin Shkreli.

Where I do see value expanding is IRL. As digital channels get flooded, the premium on in-person events should continue to accrue. Concerts, VIP, backstage, after parties - these are all events with limited space and fleeting availability. The more lux experiences are defined by exclusivity, but the broader access is defined by community. The latter aligns fans with others who enjoy their favorite artist. It leverages moments that can bring people together through shared and aligning interests, which should only gain value in a world plagued by metaversal loneliness.

I’m not the only one speculating on this thesis. There’s a great deal of money changing hands as billionaires and tech titans scramble to invest in everything from live sporting events to theater. There’s a very specific reason that the NFL is able to command so much for its broadcasting rights; live, human sport is generated in real time and unmediated by layers of slop. Reminiscing about your favorite teams still stands as one of the simplest in group / out group activities around half of America’s water coolers, and ‘catching the game last night’ generates weekly drama for its followers. Similar ideas could also explain the virality of events like Austin’s Morning Spins, and the broader wave of coffee shop DJ popups that have gained so much popularity with millennials. It’s something to do together, cultivating some modicum of community, limited by space, time, and capacity. In cyberspace, something like Twitch may occupy a more valuable niche with its primary focus on livestreaming.

Furthermore, IRL events may gain an additional premium as our virtual world continues to splinter. The internet democratized fame as it unlocked the long tail for a generation of creators. Suddenly, micro celebrity was possible and hyper niche creators could leverage social networks with infinitely larger addressable audiences to connect with their 1,000 true fans. That accessibility of culture fragmented our collective experience – your algorithm differs from my algorithm in our mediated consumption of the web. A great example of these cracks is streaming, where networks used to be able to produce monolithic hits that would capture our collective conscience (ex: The Sopranos, The Wire era) in a way that simply doesn’t exist within the multi-platform streaming wars (Apple’s Severance vs. Netflix’s Stranger Things vs. Hulu’s The Bear). AI is poised to accelerate this phenomenon as content will be increasingly hyper-personalized and algorithms will now be pitting infinitely produced AI creation against limited pools of human eyes.

The battle for our attention is as fierce as it’s ever been, and the insurgency of artificial contenders is set to intensify it. Today, video games compete with streaming competes with short form for the precious remaining hours of your life. Tomorrow, agentic apps, generative media, and AI-native hardware will enter the arena. The Slop Wars have arrived, and it’s an open question as to what exactly the collateral damage will be. I’m sure my playlists will continue to accrue beats brewed in the alchemy of algorithms, but maybe I’ll just be in hog heaven.

Anyway, here’s my sister’s reaction:

- 🍋

I got a little meme heavy this week.. My favorite Angel investment, Memelord.com just launched their newest version + mobile app, and it absolutely rocks. So snappy, so clean. If you’re in the marketing, or laughing, game… check it.

If you’d like more of my weekly musings, subscribe below:

Subscribe now

A year ago, I wrote a little riff: Spotif[AI] Between a Rock & a Hard Place.

I was mulling over the incoming onslaught of generative music, and I thought that Spotify was in a tough spot. My logic was more or less as follows:

• Integrating AI-generated music, either from users or generated from a subscriber’s taste, would help shield Spotify from disruption

• But they couldn’t do this because it would alienate the core of the business model (artist revolt, copyright questions, etc.)

• So, they’d be forced into a structural position where they’d have to wait for disruption and then overpay for whichever startup started to hit as the zeitgeist became more amenable to AI creation and/or generative music

• Hence, they didn’t have a great wedge for AI, and had only really launched some lackluster features (AI DJ, thumbnail image generation, etc.)

The piece was a five minute blast about one aspect of Spotify’s business, but I really underappreciated a couple of things.

First, the value of music as a cultural and collective experience. Sure, there will be some generative music, and we’ll get a huge AI artist sooner or later, but the value of these things to the consumer is the shared access to music we all love. Maybe I’m happy to have my elevator music personally generated on the fly for when I’m working, but the bands I adore are usually linked to experiences with others and the joy of sharing great tracks. Second, it overlooked the other ways that AI stands to enhance Spotify’s business and actually strengthen its moat.

My original framing came through the lens of music creation. I’d been writing about platforms like Roblox and the vibe coding startups, and how consumer technology broadly arcs toward accessibility. Roblox allowed anyone to create games, Uber gave everyone a personal driver, and Ableton allowed a kid in his bedroom to produce grammy caliber tracks. I took these ideas, and extrapolated to the music industry. Being a Spotify power user, I thought “Well, if there are far more creators, why wouldn’t you want to own that whole flow?” A bit of vertical integration a la Replit. Where vibe coding platforms own the creation and extract additional value through complementary services (hosting, deployment, etc.), Spotify could own the full stack from creation through distribution all in one place.

Kind of a fun idea, but misguided.

Spotify is an aggregator, and leveraging AI will serve as a sustaining innovation to their model more than a disruption. The change on the supply side actually serves to strengthen their position while simultaneously improving the value that the platform itself can create for consumers on the demand side as well.

Daniel Ek’s brilliance founding Spotify was recognizing and solving the music industry bloodbath that internet piracy had created:

One of the core insights for aggregators of that era was the structural shift that internet-native business models created. Suddenly distributors could be squashed by whichever monolith controlled demand. In music’s case, there was a hair-on-fire problem where the internet suddenly upended your packaging and distribution monopoly (CDs, supply chain, radio broadcasting) without providing the guardrails to protect your IP (see: Napster, Limewire).

Spotify stepped in to aggregate the demand, yes, but it also had a unique opportunity to wrangle highly consolidated supply from just a few dominant record labels. Then, sitting in the middle, the supply side could be delivered through a frictionless enough user experience to make its simplicity more attractive than blowing up your family’s computer with Limewire. Amazing timing.

This seed of a service grew and grew and grew to where we are today. I’m not deep enough in the industry to debate the merits of artist compensation via Spotify/Apple’s duopoly, but I can say that they’ve empowered global distribution on a scale that would previously have been impossible for any but the largest artists. If you create a song in a weekend, you can have access to 751 million Spotify listeners around the globe on Monday.

But if you’re one of those 751 million users in 2026, why is Spotify so valuable?

Sure, a decade ago Spotify was the world’s audio library, but today that’s been semi-commoditized. I can go to Apple Music or Youtube and probably find 99% of what I’d need. Apple’s leaned a bit more into exclusives than Spotify (DJ sets and stuff), and Spotify’s locked down a few select podcasts (a boon for their margins, which are constantly squeezed by record labels)... But for all intents and purposes the difference is negligible for the vast majority of users.

What’s Spotify’s really got is Wrapped:

A little tongue in cheek, but what Wrapped really represents is the marketing manifestation of their uniquely massive data lake. Spotify knows what I like, when I listen, and how I’d like to discover new music. Spotify sits in my pocket as the aggregation and discernment layer for my musical experience, fortified with a decade of my personalized taste.

I think the scale of their user base and the depth of their data expands the company’s moat in the age of AI. And that was my mistake in examining AI-powered disruption for Spotify exclusively through the supply side.

What smacked me in the face was Ben Thompson’s simple analogy:

Spotify doesn’t sell guitars.

“In terms of net new music, there are tons of companies that allow you to create music using AI. But that’s not where the music breaks. That music, if it breaks on Spotify. That’s where charts, that’s where the culture moment is. So we feel very comfortable about that position.” - Söderström 4Q25 Earnings Call

72 months ago, the world was already awash in human-generated music slop. No Suno AI needed. The 1,000x explosion of UGC should really just benefit the platform, as they remain the source that knows me best and can feed me bangers accordingly.

I also underappreciated the opportunities with their current AI products:

“More recently, we also launched Prompted Playlist, a new tool that has instantly taken off with power users. So if Interactive DJ is the chat interface to Spotify, where you can talk casually, Prompted Playlist is the deep research mode of Spotify. It lets you describe and set rules for your own personalized playlists, literally writing your own algorithm. It taps into your entire Spotify listening history, reflecting not just current obsessions, but the full arc of your music taste and integrates up to the minute culture pulled from the Internet.” - Söderström, 4Q25 Earnings Call

I fell for a little trick that lots of normies experienced – I played with their AI twelve months ago, and it was rather lackluster. It’s like your uncle who tried ChatGPT two years back, and concluded that it was an error-ridden Google knock-off. But the rate at which these things improve continues to astound (see: playing with Replit even 6mo ago vs. today), and the value of these features can actually be extracted by Spotify with only a fraction of their user base trying them. The company can pull utilization and consumer patterns out from the power users, and use this data to inform the models that are driving discovery elsewhere.

Another non-obvious example of Spotify’s unique and massive dataset comes down to vibe curation. They’re able to tailor music in real-time to an ever-evolving cultural zeitgeist.

For example, consider workout music.

I hadn’t really thought about just how idiosyncratic and ephemeral something like workout music is. I’m a 1% power user of Spotify. I DJ occasionally. I like to think I have a deep obsession with the ✨vibe ✨ that music creates. And I kind of missed how hard it must be from a data perspective to dial in these intangible kind of ideas.

The definition of “workout playlist” is constantly evolving across different people, ages, cultures, and time periods in a way that Spotify is uniquely qualified to capitalize on. My taste for workout music today is different than yours today is different than mine a year ago. Yet Spotify generally nails it.

This is one of the elements where TikTok has been insanely powerful for me (see: TikTok’s Magical Mousetrap). Its ability to dial in your likes for the algorithm broadly, and surface bangers for me specifically is astounding. But TikTok remains a video-first platform with music intertwined, not the home of all audio the way Spotify aspires.

Sure, the platform will need to prepare for a deluge of AI music-slop, but that’s actually where they’re already great– teasing out the tracks that their users want to hear. They don’t sell guitars and they definitely don’t need to.

Anyway, I guess I’m long Spotify in the forthcoming slopmaggedon.

- 🍋

Subscribe now

Man has been imagining the future of farming since the agricultural revolution. For as long as he’s had to break his back toiling away in the fields, mankind has been pushing the limitations of nature’s bounty. More for less, by any means necessary. What if we rotated crops? What if we used fertilizer? How can we escape the whims of the gods- the heat, the cold, the drought, the blight, the pests- in securing our harvest?

Given humanity’s propensity to multiply like rabbits, food security has been an ever-present concern for most of the global population throughout history. Agriculture has undergone its fair share of revolutions along the way, each pushing food production out just ahead of our exponentially growing mouths to feed:

We’ve gone from planting seeds by hand to reaping with mechanized beasts, all the way through playing god and editing DNA. With each turn of the screw, humanity has been able to eke out more crops per square foot, bolstered by the expansion of arable acreage. But for how long?

Nature has limitations. Although we’ve mastered biology so far, there are pressing concerns when it comes to food security. If not topsoil degradation, then water constraints or limitations simply on available land:

A number of solutions are pushing the boundaries of how we exploit the earth to keep our bellies full. Some like Rainmaker are betting that altering the weather will end droughts. Others are working with biology and chemistry to optimize GMOs and fertilizers. The jury’s still out on Rainmaker, and the proof has been in the pudding for our biological approaches, but another major front for innovation has been the perpetual mirage of upending our archaic agricultural processes altogether. A cyberpunk vision of scifi skyscrapers filled with greenery galore delivering precision foodstuffs with clockwork consistency. I’m talking, of course, about vertical farming.

Overview & History

Growing plants skyward is nothing new. People have been stacking greenery since the days of antiquity (see: Hanging Gardens of Bablyon), but our modern idea of vertical farming is largely attributed to Columbia University professor Dickson Despommier. In the late 90’s, he and his students were grappling with the problem of providing fresh food in dense urban environments like NYC and converged on the idea of skyscraper-scale farms. The promises of this revolutionary approach to agriculture were many: co-located food production, reduced waste, seasonal consistency, water efficiency, and the list goes on. The idea then picked up steam in the late aughts and early teens, driven by global warming fears and subsidized by low interest rates.

Modern vertical farming is the practice of growing crops in stackable, modular structures within a controlled environment, often leveraging soilless farming techniques. What this usually means in practice is giant warehouses with trays upon trays of vegetation, typically microgreens or berries, piled sky-high and illuminated by magenta LEDs. As opposed to greenhouses, which use some sunlight and are mildly climate controlled, vertical farms are sealed environments that control everything from light and temperature down to humidity and CO2 levels.

Theoretically, the advantages are many:

• Resource Efficiency: Most vertical farms use hydroponics, growing their crops in mineral rich water solutions (soilless) that deliver nutrients and oxygen directly to the plants’ roots. These systems can reduce water use vs. traditional ag by up to 90% by eliminating run off and recirculating water as it flows down the stacked trays.

• Production Consistency: Crops can be grown year-round, regardless of season and climate. Plants like spinach can produce as many as 12 harvests vs. 1-2 traditionally.

• Improved yields: Precise control of inputs (cutting edge HVAC systems, elevated CO2 levels, regulated watering and nutrients) in conjunction with the density of the building footprints can produce up to 40x more food per square meter* than traditional farms.

• Inherently Organic: Sealed growing environments allow farms to produce their crops without using pesticides, which reduces health concerns as well as pollution and run-off externalities.

• Co-Location: Vertical farms can be constructed significantly closer to urban centers, thus optimizing supply chains to reduce food waste in transit as well as refrigerated transportation costs.

*Figures vary widely by crop type, etc. But most quote orders of magnitude more production per unit of land, with some like Wheat potentially producing hundred fold improvements.

It’s the holy grail of food resilience.

So, why hasn’t it worked?

Why Hasn’t It Worked?

Vertical farming is kinda like nuclear fusion - the tech’s been ‘just five years out’ for twenty years and counting. It’s a classic venture capital bull trap. Every few years, investors find it interesting, everyone gets hyped about the promise, and then mountains of capital proceed to be torched in an attempt to will farms of the future into existence:

The most recent cycle of vertical farming startups received boatloads of investment in the last decade, with everyone from Bill Gates and Jeff Bezos to leading VCs like General Catalyst and Tiger Global backing up the truck. The promise was great and the money was cheap. The green wave and environmental fears were in full effect, then global supply chains were rocked by COVID and suddenly food resilience was the topic of the hour. The valuations grew along with the produce. The future was here!

Then reality caught up.

Vertical Farming could be pointed to as a ZIRP poster-child. When the money was cheap, it flooded into a capital intensive sector that offered great promise. The biggest names like Plenty, Appharvest, and Bowery Farming each raised hundreds of millions of dollars to sprout vertical farms across the globe, but scaled facilities before their business models were fundamentally sound. In classic Silicon Valley fashion, it was easy to grow when you were selling a dollar for $0.75, subsidizing every head of lettuce with hundreds of venture dollars.

The primary challenge was ultimately cost structure.

On the front end, you’re outfitting state of the art facilities with precision growing equipment, attempting to leverage autonomy as much as possible. This requires massive capital expenditures. For context, industry estimates from a SPAC’d Canadian vertical farmer (Beyond Farming - now Sprout AI. Because, of course) put capex costs per m³ of growing space to have significantly exceeded $3/m³ for some of the largest farms in the last cycle:

This vs. tax advantaged Kansas ag acreage that might run you $1.43 per square meter (~$5,800 per acre). Not really apples to oranges, ignoring equipment costs etc. for traditional farmland, but you get what I mean.

So, you’re starting deep in the hole just to get your racks off the ground, but then what really kills you are operating expenses. Namely, power.

Farming is an energy intensive business in its own right, but that’s fine when you have the world’s most powerful generator (the sun) free for your plants and just need a little diesel to fuel your tractor.. Once you enclose the space and have to light every hour of the growth cycle, 24/7/365, with high-end LEDs, the cost of power begins to become prohibitive. You load on top of this the ongoing power demands of an incredibly sophisticated HVAC system which is regulating humidity levels and temperature with high precision. Not to mention your ever-circulating irrigation system. Then you swap cheap migrant labor for sophisticated, high dollar robots and engineers, and pray you can automate your way to efficiency at scale. Tough.

Lighting and climate control are the real killers:

It’s a bit of a compounding problem too – you’re keeping rows upon rows of crops illuminated around the clock, but the bulbs themselves are producing heat which your HVAC system has to dissipate in order to maintain ideal temperatures. That’s in addition to combatting external temperatures, which can be particularly problematic as the most attractive locations for vertical farms tend to be the farthest from existing farmlands (ie: less moderate climates).

This translates to expensive crops:

You simply can’t deliver a cost competitive head of lettuce for $2 when the energy alone to produce it is bleeding you $3.16.

Then there’s a challenge in the product itself.

You’re producing produce.

These startups were selling an undifferentiated product to an extremely price sensitive consumer in an arena where brand name means nothing. A carrot is a carrot is a carrot. So, although you can achieve massive production gains in some crops like wheat (up to 200x in some studies!), this simply doesn’t cut it when it’s so cheap to produce the old fashioned way. Crops like cereals tend to be extremely low margin as is, and are often subsidized by governments to ensure their production.

Hence, companies tested different products and largely converged on higher margin, space efficient crops like microgreens or berries. A particularly interesting example of this is Oishii, which began with its Omakase Japanese Strawberries. These berries promise a more vibrant flavor and texture than their industrially grown counterparts. The startup’s initial product found favor with elite restaurants with deeper pocketbooks who were willing to pay up for consistent product on a consistent timeframe, and Oishii has since expanded its product portfolio to other berries and even tomatoes. This Tesla Roadster-like strategy, starting at the high end with a differentiated product and moving down the curve with scale, seems to have worked better than many of Oishii’s restructured competitors who went directly after commodity markets like lettuce or spinach.

The industry has stabilized on the back of its mass extinction event of the last few years. The remaining players have slashed their footprints, tightened their product focus, and right-sized their balance sheets for the most part. Given the state of play today, with vertical farming representing a miniscule piece of our global food production, it made me wonder: were the investors wrong or early?

Could This Time Be Different?

Vertical farming could be a textbook example of the Gartner Hype Cycle:

There was much promise, a massive wash out, and now I’m left speculating whether the sector might be entering its slope of enlightenment. “This time it’s different” are said to be the four most dangerous words in finance, but a few factors lead me to think we could be ripe for a breakout.

LED & HVAC Costs:

Light-emitting diodes (LEDs) are the most common type of lightbulb today, and have continued steadily marching down the cost curve since vertical farms were first getting attention in the late aughts. LEDs have followed their own version of Moore’s Law called Haitz’s Law, which states: “... that every decade, the cost per lumen (unit of useful light emitted) falls by a factor of 10, and the amount of light generated per LED package increases by a factor of 20, for a given wavelength (color) of light.”

LEDs have massively improved over the last decade. By 2030, they’re projected to have improved their efficacy (lumens per watt - energy efficiency) by ~90% from 2010. This was occurring concurrently to the average retail price falling precipitously:

These efficiencies could compound as they’d reduce upfront capex initially, and also improve ongoing energy costs on the HVAC side with their reduced heat output.

Unfortunately, on the latter side of that equation, HVAC systems are still largely physics constrained as vertical farming requires energy intensive dehumidification processes which don’t seem to have experienced the same step-change improvements in the last decade.

Robotics, Sensors, Intelligence

Similarly, while the vast majority of vertical farms were getting wrecked by their expensive cost structures, the requisite components of automation continued to advance. Robotics has made leaps and bounds in both quality and cost of computer vision systems, the types of which are needed to monitor your crops, as well as the general automation needed for planting, picking, and other activities within your warehouse.

Orchard AI is a prime example of these kinds of breakthroughs applied to the ag space already. They’ve invented a tractor-mountable camera system with the ability to capture 100 images per second, recording data about every fruit on every tree within a farmer’s orchard. Placing hardware with that level of fidelity (camera quality, edge compute) at an attractive price point to farmers was an economic impossibility a decade ago, let alone processing the terabytes of images at scale on the back end. These same hardware improvements and AI capabilities could translate back into the controlled environment agriculture arena as well.

On the labor automation front, we’ve also seen strides. ‘Embodied intelligence’ has become all the rage as investors seek the next leg up of the artificial intelligence boom, and those investments have driven costs down rapidly:

Simultaneously, dexterity has continued to improve (much needed to pick delicate little berries):

These two components alone have made me wonder if a blank slate startup, unbound by technical debt like existing incumbents, could sidestep some of the pitfalls of the previous generation and chart a more expedient path to scaled profitability.

Unfortunately though, physics is undefeated and the largest cost driver seems to be going against you.

Energy:

Energy prices continue to climb. Feeding the insatiable appetite of our digital gigabrians has come at the cost of maxing our grid and squeezing our pocketbooks. This is a particularly thorny problem if you’re a vertical farm and your primary cost driver is the price of a kilowatt.

The irony of this challenge is that it’s precisely what led me to explore vertical farming in the first place. I’ve been contemplating the second order effects of energy superabundance. My thinking gets speculative here, but allow me to explain.

I previously wrote about the race to ooch our way up the Kardeshev Scale, and master energy on a societal scale. Well, now it really feels like we’ve undergone a generational vibe shift with regard to power production. Across the aisle, there has been unanimous recognition of its underlying importance in our competition with China, and how sorely we’re losing that race:

These economic and competitive pressures have catalyzed a number of bold policy moves. Everything from abducting heads of state to accelerating the nuclear renaissance. Done responsibly, dramatically increasing America’s energy generation is a net positive for our society. That said, the types of investments that actually make a dent are extremely capital intensive and extremely long-dated. Even with the SMR race to reach Criticality by 2026, home town heroes like Aalo still won’t actually be producing reactors at grid scale for a number of years.

Which leads me to believe that there’s a very real possibility that the AI revolution follows a similar pattern to the dotcom bubble with regard to infrastructural overbuilding.

The Dotcom boom pulled insane amounts of investment into telecom infrastructure, which was subsequently underutilized and crushed the sector as internet demand fell off when the bubble burst. Similarly, I think that a breakthrough in pre-/post- training efficiency or chip architecture on the AI front (and/or any other black swan into the trough of disillusionment) could potentially leave the U.S. in a place where we’ve massively overbuilt power infrastructure. Given the titanic nature of these types of projects, it could easily result in a long period of very cheap energy.

By way of example, average U.S. and Chinese industrial energy pricing today sits around 8-9¢/kWh, and then startups like Aalo are ambitiously targeting sub-3¢/kWh:

It’s possible that a vertical farming startup founded in 2026 could be reaching scale just as energy prices are falling off a cliff. A wave of cheaper supply coming online concurrent to declining demand could be a great place for a vertical farm to flourish.

Again, speculative but not crazy.

My other scifi thought was really the colocation of SMRs with vertical farms following the same logic we’re applying to datacenters. If ever there were a piece of infrastructure that needed steady baseload power, a consistently humming vertical farm could very much be it.

Conclusion

Vertical farming seems like an interesting, out of favor sector that could potentially produce a venture scale winner. A generation of investors have been burnt by its prospects, and hence the pricing could be particularly attractive on the right deal. Perhaps there’s a contrarian take where the cost physics have shifted beneath the industry’s feet in such a way that a winner could bloom at scale.

History doesn’t repeat itself, but it does often rhyme. Pets.com is frequently named in the same breath as dotcom exuberance, yet the explosion of the internet gave way to unicorn Chewy not ten years later. Same shit, different decades. Perhaps vertical farming could be the ZIRP era equivalent.

But hey, maybe the food problem works itself out, regardless. Perhaps population collapse really is imminent and the Cassandras out in the Valley prove to be right. Were that the case, I still see this technology as a relevant component of domestic food security in an increasingly isolationist geopolitical world.

Plus, how else does Elon think we’re going to grow all that food on Mars?

I’ve seen Bladerunner. Grub greenhouses and vertical farms.

- 🍋

P.S. Big thanks to Jack Oslan, a Co-Founder of Plenty, for graciously talking Vertical Farming with me for a couple of hours.

Subscribe now

Thinking about AI, two quotes keep swirling around my mind:

“Our job is not to see the future, it’s to see the present more clearly.” - Matt Cohler, Benchmark

“The future is already here, it’s just not evenly distributed.” - William Gibson

It’s apparent to anyone playing with these tools that AI is going to change the future. For most of us outside of the research sphere, this idea became obvious with the launch of ChatGPT a few years ago. The models are exponentially improving, the talent wars are raging, and the launch videos are hyping, yet for all the money poured down the AI funnel, it still feels a bit like agentic software is still in its Clippy era:

For the average normie, AI’s effects have largely been felt through:

• Marginally improved search functionality - like Google with a bit better results

• Better chatbots on corporate websites - many of which still provide nonanswers

• AI slop content - that’s already brainwashing the most susceptible boomers

For those of us a bit deeper in the sauce, we’re starting to see its effects more clearly. Just ask anyone you know developing software:

Playing with some of the more agentic tools, I feel like I’m getting a very clear glimpse at the unfolding present. I certainly can’t see the future, but a couple avenues of the timeline that interest me are AI’s GUI moment and its first real hardware hit.

I was one of the judges for Jason Calacanis’s LAUNCH demo day a couple weeks ago, and I caught the tail end of a talk given by Superhuman cofounder Vivek Sodera. He was outlining a broad thesis that the future of software is going to be far more fluid. The idea is that our interactions with computers are about to change radically through a more amorphous kind of ‘just in time’ software experience. LLMs will produce bespoke or tailored software solutions as the need arises, then compartmentalize the programs in a way that’s very different from our laborious download+click+search+type+etc. workflow of today. I’d had inklings of this idea percolating, but his talk helped to firm up some of my thoughts. Bear with me though, as they’re still a bit gelatinous.

It made me think of computing’s GUI moment.

In the nascent days of software, computers were accessed through a combination of command prompts and coding, where the user was essentially typing very specific functions into a computer to output the desired actions.

This process was extremely high friction. It required specialized knowledge in order to invoke basic actions from a computer, which was unintuitive and cumbersome. Utilizing a computer in the pre-GUI era was essentially coding in its rawest form, inputting the specific commands, step-by-step, that we now wrap into even the most basic software today. The slightest syntax error could torpedo your entire workflow. It was a clear problem that early technologists had set out to fix.

Famously, Xerox PARC, the R&D hub of the early computing pioneer, invented a version of the graphical user interface. It was comprised of a point-and-click functionality that allowed users to command their computer in a more intuitive way, utilizing a mouse and visual representations of files, functions, and programs. This was a breakthrough in user experience, but the stodgy tech titan lacked the vision to really commercialize it.

Through one of fate’s great chance encounters, Steve Jobs was shown a demo of the interface while touring the Xerox campus and immediately grasped the magnitude of what they had invented. He would later say that it was like a veil had been lifted for him, and that Xerox could have owned the entire computer industry had they realized their brilliance. Apple quickly rolled this functionality into their Lisa product, and helped to birth the user interface that defines computing today.

The Graphical User Interface made interacting with technology so simple your grandma could do it. It was more intuitive, more natural, and more accessible in a way that broadened the addressable audience for computers by an order of magnitude. In conjunction with more powerful chips and a burgeoning software development industry, the GUI helped to power the computer’s broader adoption as a business tool in corporate America. The revolution had begun.

I think that artificial intelligence may be the next turning of this screw. There’s a direct impact on software itself, then there’s the broader redefining of computer interfacing as a whole.

For the former, I feel like vibe coding has given me a taste of this narrowly distributed future. These tools are accessible, but not widely known, and limited by the distribution of the startups themselves (though this is growing exponentially). The way that basic interactions with the computers of the 70’s required a nuanced understanding of command line prompting, software development today has been a gatekept profession. In order to speak computer in 2010, you still needed to know a coding language of some sort (C++, Java, etc.). The friction of learning this skillset helped to create a hacker class that was extremely well compensated for its expertise. Computers and the internet created infinite leverage, but it was a select few within the broader population that could really run with this force multiplier. Then came no-code tools as an attempt to close that gap: a generation of simplified software that could empower the normies. This went a long way, but didn’t quite blow off the doors the way agentic coding has…

Now your grandma can create an alarm clock app with natural language. No coding, no hosting, no complex back end. Replit’s got her covered. Software will become content (a drum I keep beating), and the accompanying economics will rapidly degrade.

Remember the Coldplay Kiss Cam scandal? A guy vibe coded that into a game within a day:

And this extends beyond memetic games. The nature of industrial-grade software development is transforming as well:

One exceptional software architect with an army of agents can ship with the veracity of a dozen Comp. Sci. interns. Software development isn’t immune to power laws.

This upheaval also interplays with the next generation of AI-native programs and the operating systems that will access them.

Currently, most of our AI-based interactions happen via an intermediary native to the program or through a third party app itself. When I use AI, I’m going to the actual ChatGPT site or I’m interacting with the Help Bot on the Chase page. These are interesting and have their clear use cases, but I think the next leg up could be a far more dramatic reimagining of software.

A stupid example:

I have a long standing beef with the fact that neither Google Docs nor Google Sheets have a native Dark Mode. It’s 2025, we’re inventing god intelligences to annihilate humanity in the next few years, and Google still hasn’t added these basic features. Wild. This is an imminently solvable problem. Yes, I know chrome extensions like Dark Reader exist, but in our imagined future this fix would be merely a prompt away.

Where that sits in the stack, I’m still unsure, but there’s a few ways it could play:

App level: “Hey Gemini? You’re scorching my corneas. Please implement dark mode into this web page for me and add a button at the end of the toolbar to toggle this feature.”

YC has famously said “Make something people want.” What if they could make it themselves within your software? No more help tickets and feature requests… Software could be built to give your users the reins, empowering decentralized product development while also driving tighter feedback cycles, enabling builders to push some amount of their roadmap ideation directly into the hands of users. Or perhaps it’s a world where the software that each of us uses is as customized as the highly personalized feeds that dominate our social media platforms.

Browser level: “Hey Chrome, as I surf the web 🏄‍♀️, please keep my precious peepers from getting sauteed. Any web page that has primarily white background features or light/bright colors dominating the field of view, please flip to be inverted (dark mode). Do not perform this color inversion on images, icons, or logos on the pages themselves.”

Of course this is far out today (and prohibitively expensive in terms of compute, I’m sure), but it might not be an unreasonable future. Though, the first two examples above still feel a bit skeuomorphic. They’re merely overlaying kinda new ideas on top of legacy tech.

More radical reimaginings could manifest at the operating system or hardware levels.

Back to the GUI idea, the entire interface with our laptop should be able to evolve. Operating systems will soon host native agents (think Microsoft’s Clippy as your laptop copilot) as well as advanced voice processing software. One of the breakthroughs with LLMs is natural language processing that could soon allow us to navigate our computers at the speed of sound, dictating commands as they come to mind. You could even bolster its effectiveness by leveraging something like the cutting edge eye tracking capabilities that Apple honed while developing the Vision Pro. A more capable computer could be reading your eye movements, tracking your voice, and whizzing away through emails like never before (the most depressing use case possible).

Think about it- Why the hell am I clicking into Twitter, clicking Tweet, typing my thought, then clicking the upload button, navigating 10 folders, finding the meme I’d like to add, then uploading it, then clicking send?

Typing the eight steps it took me to post that shitty meme above already makes it feel like an incredibly archaic workflow. The way VC’s dictate ideas to their ghostwriters, I ought to be able to have my Clippy whip it up in a fraction of the time.

Then I take this to its logical conclusion, which is a reinvention of the hardware itself. Models today are massive, so a lot of the AI-native hardware experiments like Friend or Limitless are really more like edge nodes, with the heavy compute being performed elsewhere. This may not always be the case. Models will continue to get better, and eventually they will get smaller and more efficient as architecture gets optimized and additional breakthroughs occur. Additionally, if Moore’s law can continue its steady march forward, these forces will converge to produce incredibly powerful agents in your pocket.

We’re already seeing tech giants like Meta going all in on their estimation that the smart phone will eventually be supplanted by a more intuitive form factor:

Mark Zuckerberg certainly seems to believe that it’s some kind of headwear. An agent in your pocket will lack the kind of rich context that a camera at eye level will be able to ingest. Plus, a heads up display in your field of view may prove to have more utility than pulling out your phone every thirty seconds. Other tech giants share this thinking, with players like Anduril going deep on AR/XR technology for the warfighter with their Eagle Eye augmentation platform. Palmer Lucky’s looking to create American Technomancers and envisions a future where “…you’re going to see a head-mounted display of some kind on basically every single head in the DoD.” (Core Memory). If the hardware can get small, light, and powerful enough, the aperture for how we infuse our life with technology will dramatically change.

Of course, this phase shift won’t be all or nothing, and no change happens overnight. For all my enthusiasm, I’ve dramatically oversimplified quite a bit of the messy middle ground between an idealized future and chatbots spewing corporate slop today. Even the most tech forward companies are seeing their AI implementations underwhelm.

That said, these evolutions feel inevitable. The GUI was a step function change for our relationship with software. Its effect was profound because it abstracted away the dull and dirty parts of navigating technology. Commoditized intelligence has the opportunity to abstract away so much more in ways that we’re only beginning to glimpse today. The Clippys of the past were a meme for their friction, but the Cortanas of the future might just be the pilots we need.

- 🍋

Subscribe now

I love learning something new.

New skill, new hobby, new book, new tech– let me at ‘em. This voracious curiosity is probably why I’m a serial generalist… and perhaps why I’m an investor. Back when I was in public markets, I used to say that “Investing is the greatest job in the world. I get paid to learn.” Now that I’m in venture the privilege is one better, but the point still stands.

One of the beautiful things about climbing any skill tree is getting a taste for just how incredibly masterful the people in the highest branches are.

This week’s slice of humble pie has come from vibe coding. Replit released V3 of their Agent, and I felt that it would negligent for the technologist in me not to give it a whirl. I’ve previously vibe coded a project or two, so I was excited to see how much better the tech had gotten. I thought I’d start with something dumb to keep it very simple.

The problem: my buddy runs an annual pick ‘em league every NFL season, and he organizes this fun little tradition through the world’s jankiest app:

I figured that one upping this Windows 94-looking app would be an easy side project.

I’ll spare you the gory details on the logistics of the league (and how I got labeled ‘Lil Gup’ 🐟), but suffice to say that it’s a relatively simple idea: populate some weekly games, pull some Vegas spreads, add some snappy point calculations. Let the shit talking commence.

The only problem is, I am utterly untechnical. The extent of my coding expertise is limited to MySpace html in middle school, and yelling at Replit over and over to brute force a GPT-wrapper that would teach me rhetoric…

So, of course, I started prompting gibberish and crafting the skeleton of a crappy MVP:

Duking it out with Replit half the weekend to bring my little vision to life reminded of a classic Ira Glass clip:

“Nobody tells people who are beginners… All of us who do creative work, we get into it because we have good taste. Do you know what I mean?

You want to make TV because you love TV. There’s stuff that you just love. Ok.

So you’ve got really good taste, and you get into this thing that- and I don’t know how to describe it, but it’s like there’s a gap. That for the first couple years that you’re making stuff… what you’re making isn’t so good. Ok? It’s not that great. It’s really not that great.

It’s trying to be good. It has ambition to be good, but it’s not quite that good.

But your taste, the thing that got you into the game, your taste is still killer. And your taste is good enough that you can tell that what you’re making is kind of a disappointment to you. You know what I mean? Like, you can tell that it’s still kind of crappy.

A lot of people never get past that phase. A lot of people at that point, they quit.

And the thing that I would just like say to you with all my heart is that most everybody I know who does interesting creative work- they went through a phase of years where they had really good taste, they could tell that what they were making wasn’t as good as they wanted it to be, they knew it fell short… it didn’t have the special thing that we wanted it to have.

The thing that I would say to you is that everybody goes through that. And for you to go through it, if you’re going through it right now, if you’re just getting out of that phase- or if you’re just starting off and you’re entering into that phase, you gotta know it’s totally normal and the most important possible thing you can do is do a lot of work.

Do a huge volume of work. Put yourself on a deadline so that every week or every month, you know you’re going to finish one story, you know what I mean? Whatever it’s going to be, but you create the deadline. It’s best if you have somebody who’s waiting for work from you, somebody who’s expecting work from you. Even if it’s not somebody who pays you, but that you’re in a situation where you have to turn out the work.

It’s only by actually going through a volume of work that you’re actually going to catch up and close that gap, and the work you’re making will be as good as your ambitions.” - Ira Glass on Storytelling

One of my favorite videos on skill from a legendary creator. It’s a resounding endorsement of the volume required to be great.

It’s a truism that I’m constantly relearning. I’m an ambitious optimist with a penchant for trying whatever I’m really into lately, so I get my fair share of reminders. This week is some software development, historically that’s been music (playing, producing, DJing). I think there are parallels in both of these fields that really prove the point.

At the highest levels, both software development and music production are manifestations of creative taste, forged through thousands of hours of practice. Done well, the outputs of both belie their complexity. Beautiful apps and bangin’ tracks both appear deceptively polished and devilishly simple.

Calvin Harris is a great example of this from the world of dance music. I’m not even his biggest fan, but I have an immense respect for the sheen of his mega hits. The man can brew a banger. He’s mastered the formula of surfacing once every few years, finding the hottest female pop star of the hour, tossing her vocals over a dance beat, and minting a certified heater:

• Outside (ft. Ellie Goulding), 2014 - 1.6B streams

• This is What We Came For (ft. Rihanna), 2016 - 2.1B streams

• One Kiss (ft. Dua Lipa), 2018 - 2.5B streams

All three of these tracks are commercially immaculate. The cadence, the sparkle, the sex appeal of each combined to produce a certified summer hit, year after year. The tracks seem so simple- formulaic, even.. Like modern art, you think to yourself “Hell, I could do that.”

Then you download Ableton.

The hours that went into producing those perfect three and a half minute bops are mind boggling. Between recording, mixing, mastering, and all that jazz, the devil hides deep in the details of what it takes to make a track sing. The arc of accessibility bends toward bringing down the underlying costs and simplifying some of the inputs, but the hours to polish a craft still remain.

We hear this level of excellence so frequently that it’s easy to forget how amazing it really is.

The actual playing of an instrument is no different. When I was last taking saxophone lessons, I remember thinking just how skilled almost every musician you see is. Even the guys in the jam band playing that dive bar on Rainey are so far beyond what modicum of skill I gathered playing for 5 years. Every DJ on every label, big and small, has put countless hours into their craft.

Software is much the same. Amazing software is forgettable. It’s sleek, it’s intuitive. It’s so natural as to be second nature. It’s the little bubble rubber banding on your iphone, so subtly satisfying.

I try to vibe code half a skeleton of a quarter of an app…

…and immediately realize I don’t understand UX/UI, data bases, APIs, authorizations and any of the other million little things that go into making crisp software. Once you start putting your money where your mouth is, you have quite a bit more empathy when X, an app running millions of actions per minute (second?) at a global scale, is down for 5 minutes.

We’re so used to everything being amazing that we’re aghast by the slightest inconveniences in experiences that are otherwise so profound as to be magical:

But that’s really the joy of being an amateur. You can once again be intimidated by the gulf between your ideas and the execution required to breathe life into them. The volume Ira mentions reminds me simply of the reps required to achieve proficiency. You look at the gear of any new hobby, from the hooks and baubles of a master fisherman to the knobs and sliders of a maestro producer, and you appreciate the elemental mastery a virtuoso commands.

Sometimes this chasm can be offputting. It’s endlessly frustrating to willingly throw yourself into the gap between your taste and your abilities. It’s demoralizing to see the road ahead. “If only I’d begun years ago..”, you might say to yourself.

As with anything worth chasing, the best time to plant a seed was 30 years ago, the second best time is now.

So, I implore you to do something hard. There is so much joy in learning something new, and we’re so fortunate to benefit from those who chase their obsessions.

- 🍋

Subscribe now

In the early days of streaming, there was once a strategic conundrum of great debate: could Netflix become cable before cable became Netflix?

Netflix was a disruptive aggregator that thrived by modularizing content while integrating production, subscriptions, and distribution:

The startup had built a platform that aggregated content from all the producers/studios, and owned the infinite distribution that the internet had enabled. This allowed Netflix to scale rapidly, reaping the decreasing CAC that comes with being an aggregator.

As the streaming giant grew, an inevitable tension began to emerge: this type of market consolidation inevitably squeezes suppliers (the studios/networks) as the aggregator A) gains buying power B) increasingly commoditizes them. Netflix was destined to bite the hand that fed; a relationship which was once complimentary for the content suppliers, gaining distribution, would eventually compress their margins. The tension between these two increased exponentially with Netflix’s share of the market.

Thus, the race was on. Could Netflix commoditize its suppliers before they worked out an alternative strategy for distribution or differentiation?

Eventually, everyone and their dog became a streaming service.. But not before Netflix was able to flex its balance sheet and backward integrate, producing its own prestige content. For a rough timeline, Netflix began streaming in 2007 and Hulu launched in 2008, but was largely ad-supported vs. subscriptions. Although there were other attempts (Amazon, etc.) along the way, the streaming wars themselves really didn’t ignite until the late teens with the launches of Disney+, Apple TV, Peacock, and HBO Max. By then, it was far too late. Netflix had been producing its own originals since ~2013, with major successes like House of Cards and Orange is the New Black before mega hits like 2015’s Narcos and Stranger Things. The platform’s leverage was established and, although the content landscape has fragmented in recent years, it still remains in pull position today.

I think we’re destined to see a similar dynamic play out today between design software and vibe coding platforms. From my personal use, I’ve been wondering if Replit eats Figma before Figma can eat Replit.

Right now, these platforms are more-or-less frenemies. Vibe coding empowers designers, so Figma is happy to integrate with whomever would like to plug in. Conversely, players like Replit are happy to tap Figma’s massive user-base of tech savvy customers to leverage distribution. Symbiotic.

But eventually, this feels like a modern scorpion and the toad situation:

Today, Figma is the bigger fish swimming in a different pond. Their target audience historically has been designers and teams building digital experiences. They’re approaching $1B ARR (still growing 46% YoY, mind you) across 13 million monthly active users. This is compared to the leading vibe coders like Replit and Lovable, who are both doing around $100M ARR on significantly smaller user bases of software developers and the ‘coding curious’. Replit claims 500K ‘professional users’ and Lovable claims 180K paying subscribers/2.3 MAU.

The scales and targets are different for now, but I believe their end goals are destined to collide. Both vibe coders and design platforms are ultimately tools for democratizing software creation. They’re similar accelerants for the arc of accessibility, an idea I previously explored. Although Figma’s bread and butter is professional designers working cross-functionally to bring ideas to life, they specifically note in their S-1 that about ⅔ of their users are non-designers. I imagine that Replit’s user split is probably in the same ballpark, with offerings for professional-grade devs but a stated goal for anyone to create software.

Just look at the company visions:

Some of it is corporate slop, but there’s a clear focal point of their mutual aims: be the go-to platform for creativity expressed through software.

Which begs the question: who will eat whom?

The ouraboros has already started to feast:

Replit has robust integrations to simply flip your wireframes from Figma into Replit’s platform, then build out the functionality via its Agents. But they’ve also bolstered the native UX functionality with design-phase planning in new projects and tools like Visual Editor, which allows you to make edits directly to the UI with real-time code updates. Conversely, Figma has always supercharged the speed to ship software with resources like Figma Dev Mode, but now they’re expanding their purview with native prototyping via Figma Make:

This convergence creates a very interesting strategic dynamic. If you’re Figma, you kind of want to do the Netflix thing. Continue aggregating a very specific and very valuable audience while commoditizing your compliments. As long as you own the distribution, you’re in a prime position to be the middle layer of creation for your users while simultaneously leveraging increasingly interchangeable AI models for the actual development. In and of itself, Replit does not build the underlying LLMs in-house, which implies that there may be an opportunity to eventually disintermediate the software creation player. If you’re Figma, you want to continue expanding your actual software development offering, leveraging whichever model as the back-end of your infra. This allows you to provide an improved user experience while capitalizing on market competition to provide the intelligence (ideally, helping to maintain your margins). If Figma Make gets sophisticated enough with its prototyping, when would I ever leave to Replit?

So, where’s that leave Replit? If you’re in their shoes, do you really want to go to war on the aesthetic front for that user-base, or is the TAM big enough simply drawing in normies like me? Hard to tell as paths converge. My journey to Figma came via my janky prototyping in Replit and a desire for more tasteful UI. But then-again, in a world where software becomes content maybe there are more attractive vectors for differentiation.

The same way that models themselves potentially get commoditized, I think that vibe coding likely does as well. Not that a huge player won’t come out of that space, but that competition will be fierce and margins likely compress. There’s already a crowded field of players coming for this crown. The Replits and Lovables of the world are trying to onboard the next generation of software developers while a score of other vibe coding platforms are going after prosumer and enterprise. All the while, the LLM developers themselves are offering ever more agentic features natively.

This is why the end platforms will be defined by more than simply code. My ChatGPT interface can rip code with the best of them today, but it lacks the context, integrations, and capabilities to deploy full stack creations. Sure, perhaps its independent agents will be able to take those steps for me eventually, but having an integrated solution on a single platform will have significantly less friction.

The lift for design platforms to integrate better models into their stack is likely lower than the technical know-how it will take to deploy full user-grade applications in a one-stop shop. Plus, providing the infrastructural back-end to enable your creators to flex up and down with your platform as they scale will be important. Replit wants to become as Shopify-like as possible, allowing builders to go from idea to deployment to 100,000 customers with their tools, all while living within Replit’s ecosystem. As we’ve seen elsewhere with the likes of Roblox, providing the tools and infrastructure for user creativity and monetization is a highly lucrative position to occupy. Perhaps the market bifurcates into code creation layers for rapid prototyping through the likes of Figma, while the Replits of the world are forced to move more vertically in order to offer a differentiated solution for their users.

That’s the current landscape.

Figma Make is a fun prototyping tool, but lacks the depth to really ‘ship to prod’. Could they build these capabilities in-house? Maybe. But it would be slow and they probably lack the real expertise to stand up the needed infrastructure. A more interesting question is whether or not they should simply buy their way into vertical integration..

Figma IPO’d about a month ago, adding ~$412M in cash to their now $1.2B warchest. They’re doing over $1B ARR at this point at 90% margins, growing almost 50% YoY. On top of that, they have about zero debt on the balance sheet. Against their $28B market cap, a material acquisition of an agentic coding platform is well within reach.

There’s precedent for such a move:

Figma would be quite a bit later to the game vs. someone like Wix who was able to scoop up Base44 for $80M (amazing buy so far), but nonetheless the acquisition route would get them up the learning curve far more rapidly. All the while, Figma’s seeing its core business attacked by such upstarts as AI-native website builder Framer ($2B val/500K MAU):

Ultimately, we are living through a stunning period of tech disruption. Figma’s future, like every incumbent on the block, will be defined by its ability to adapt to an agentic paradigm. The cost to ship is collapsing, and simple prototyping on their platform may not be enough to maintain the company’s moat around software creation.

Admittedly, it’s far too early to suggest that Figma has stumbled with its AI implementations; they’ve been generally well received. But this could be a moment where cracks begin to form. Their core business is under assault from upstart players, and complementary products are gnawing at their sides. With its recent public offering, Figma has the ability to leverage public markets in a way that its competitors cannot. If their scale and IPO have lowered the company’s cost of capital, an acquisition like Replit might just be the opportunity to eat vibe coding before the scorpions come for design.

- 🍋

Subscribe now

Injury, the bane of every athlete’s existence. Minor, major, temporary, niggling- the flavors of frustration abound.

I’ve been plagued by my own spectre recently.

A month ago, I experienced the first lower back injury of my lifting career. You learn a bit in success; you learn a lot in failure. This holds true across sport, business, and life. And, brother, the last few weeks have got me introspecting.

Injuries pain more than just the body, they become a virus of the mind. They infect your day to day, rearing their ugly head each time you feel a tinge. Every untoward motion reminds you that the machine is broken. You are broken.

Cue existential panic.

Then the months and years afterward these accumulated chinks in the armor sit like little ghouls at the back of your mind: “my trick knee, my bad shoulder, my damn back.” They’re the darker hanger-ons that accompany the pursuit of athletic achievement.

In the moment, as I am now, you’re viscerally reminded of that old truism: “A well man wants a thousand things, a sick man wants only one.” The yin to this yang is a forced reminder to be grateful. Perhaps it’s human nature to never know what you got ‘till it’s gone, but breaks like these force a little space in the routine to appreciate where you were and where you’ve been. Hell, I was once tossing around 400lb, an incredible feat for the stick boy freshman that started this journey.

Which helps you to step back- how long have you been devoted to this pursuit, and just how much of your identity is intertwined in its continuation?

In my darker moments, there’s certainly been a bit of spiral… how much of my identity, but also my lifestyle, my joy, and my personal ambition, have been intertwined in this hobby of mine? One of the many joys in life is seeing just how far you can go, whatever skill tree you pursue. Hop the plane, climb the mountain, run the marathon. In these moments of setback, you’re forced to contemplate if the ceiling of those experiences has somehow been handicapped?

These ideas extend beyond fitness, of course.

How many people define themselves by their job title? And how many are so completely rudderless if that’s torn from them?

Setbacks shake the snowglobe.

I first heard this metaphor used for neuroplasticity. A young brain is like a freshly powdered mountain, with many routes to reach an end behavior. But after a few passes down the slopes, ruts begin to form which become the default for our reactions. There was some thought that psychedelic-assisted therapies helped to shake the snowglobe and re-powder the pathways.

I’ve had a similar silver lining. My weightlifting career spans over a decade. I’ve bulked and cut and researched just about everything under the sun, but the most recent injury forced a pause for reassessment. Going to PT for the first time was the shake I needed.

The expertise of a trained professional has been a healthy reminder that you can’t do it all. The specialization of others is how we scale ourselves. As with companies, the team of you is no different. Hearing her fresh perspective, and the insights I could glean, provided new frameworks for assessing my training and my movements. We’re often so myopic in our problem-solving that we fail to examine the broader system. Problems of work and play are both relevant in that regard.

Which also made me think about pivots.

Yes, I know. Techbro at his finest. Don’t worry, I’ll close the piece with the 5 things back pain taught me about B2B sales.

Jokes aside, this unfortunate shock could be a forcing function to evolve. Just how can I train around this injury, or where else can I pour my energy? How am I fulfilling my drive for self improvement in more sustainable ways?

You probably aren’t throwing around 400lb at 40, and you probably aren’t 9-9-6’ing with a family. Your training and your career both evolve as you age; sometimes you level up, sometimes you shift tacts, and sometimes you overhaul the system altogether.

Setbacks shake the snowglobe and remind you what you’re made of. Grit was never grown on easy street.

- 🍋

Subscribe now

If we lurk the same corner of tech twitter, you probably see some techno optimist invoking the Kardashev Scale about once a week. To some, it’s become the ultimate arbiter of progress to be climbed with a religious zeal. We inhabit a godless postmodern hellscape, thus harnessing the entropy of the universe is a calling to transcend our meager toils. The idea is part and parcel of the Effective Accelerationism (e/acc) movement; a philosophy defined by “... embracing the thermodynamic imperative of the universe to accelerate progress, intelligence, and civilization” in pursuit of a “post-scarcity technological utopia”. Pretty cerebral stuff.

If e/acc is a cult, Beff Jezos is the prophet, and the Kardashev Scale is its gospel.

Simply, the Kardashev Scale is a measure of a civilization’s progress as defined by its ability to harness energy. It was first proposed in 1964 by Soviet astronomer Nikolai Kardashev as a framework in the search for other civilizations in the universe, and it’s comprised of a few broad categories:

A Type 0 civilization (added later by Carl Sagan) has harnessed about as much terrestrial energy as a few dozen campfires. Barely even to the starting line. On a galactic scale, humanity is in this neighborhood.

A Type I civilization is able to access all the energy available to its planet and store it for consumption. This could include resources like fossil fuels, wind, geothermal, and nuclear – but realistically, it means capturing all of the energy hitting a planet from its parent star (in Earth’s case, the sun), which produces orders of magnitude more than all terrestrial resources combined.

A Type II civilization can directly consume a star's energy, most likely through the use of a Dyson Sphere or some similar megastructure. This level of sophistication would imply mastery over interplanetary travel, off-world manufacturing, and engineering on a scale scarcely comprehensible to us today.

A Type III civilization is able to capture all the energy emitted by its galaxy, and every object within it, such as every star, black hole, etc. Past this point is when we’ve really entered deep sci-fi territory. If “any sufficiently advanced technology is indistinguishable from magic”, a civilization at this scale would likely look like gods to us.

Unfortunately for humanity, we haven’t even cusped level I yet. We’re barely a speck of dust on the universal stage. Most futurists place us around Type 0.7 (aka Type 0), and project that we ought to be able to climb our way to level I in the next hundred years or so. In technical terms, total human energy consumption is about 2 × 10¹³ watts, and Type I is reached at ~10¹⁶ . Each leg up thereafter goes by orders of magnitude (Type II ~10²⁶ watts, Type III ~10³⁶ watts).

Overintellectualizing aside, the Kardashev Scale is useful for its simplicity. It distills all progress down to a single tangible metric: energy consumption.

Of course, by this yardstick, we’re still practically cavemen burning twigs…

Yet with the launch of the AI arms race, this may be subject to change. The biggest AI story besides the breathtaking advancements of the tech itself has become the greed with which it demands power. The chart of the hour recently was the insane explosion of data center construction for our AI overlords (soon displacing new builds to house our corporate drones):

The second order effects of which are starting to show:

And these ideas are important. Power is important.

He who controls the compute may control destiny…

But all is lost without the power to fuel your intelligence and turn that into dexterity. We may have found a way to conjure intelligence at scale, but that’s really the combined resources of silicon and electricity. We’re making rocks speak, but you need the thinky juice to do it.. And our nearest competitor has bet accordingly:

A couple of years ago, Isaiah Taylor, the founder of nuclear company Valar Atomics, told me something that’s stuck in my head ever since:

There are only really three pillars to anything around us, as far as consumable goods. We've got energy, intelligence, and dexterity.

I would generalize “dexterity” to “action.” Everything we see around us, and will see around us in the future, is the result of the potential to do work (energy), the capacity to decide what to do and how (intelligence), and the ability to manipulate matter (action).

In economic terms, energy, intelligence, and action are strong complements in the production of anything.

And in the immortal words of Joel Spolsky, "Smart companies try to commoditize their products’ complements.”

America is, implicitly or explicitly, making a bet that whoever wins intelligence, in the form of AI, wins the future.

China is making a different bet: that for intelligence to truly matter, it needs energy and action.

If you control energy and action, making intelligence abundant strengthens your position.

After catching up to America in electricity generation in just 2010, China now generates 2.5x as much electricity as we do.

- Packy McCormick, The Electric Slide (Not Boring)

This extends beyond AI as well though, as energy accessibility is one of the defining features of advanced nations:

One can debate if the tail wags the dog above, but the one thing that isn’t up for debate is the freedom afforded by cheap energy:

“Oh, we’re running out of clean water, but it’s too expensive to desalinate!”

Not with infinite energy.

“We don’t have enough arable land!”

Scale your vertical farms and their 24/7 grow-lights.

“But these advanced industrial processes simply aren’t economic!”

Problem solved when your largest cost center goes to zero.

Many of our greatest thinkers and tinkerers recognize this. The might of American AI and domestic reindustrialization will depend in no small part upon our access to abundant energy. Investors have awoken to this truth and are pouring millions of dollars into moving the ball forward.

Hence, so many cool startups working to ooch our way up the Kardashev Scale.

Two themes in this narrative that have captured my attention are the nuclear renaissance and extraterrestrial solar.

Short term power demands are mounting. Without major breakthroughs in efficiency, the power consumed in running the world’s most advanced AI models will soon rival that of entire nations. This is compounded by the consistency with which data centers need power as well. Grid scale battery farms continue to advance, but data centers generally skew away from wind or solar toward less clean sources of power for lack of this kind of infrastructure (MIT Tech Review). But what energy source has a small footprint, low emissions, 24/7/365 output, and multiple decades of data proving its safety?

Enter: the Nuclear Renaissance.

I casually started following the radioactive vibe shift in the fall of 2023, just before Packy launched his nuke-pilled podcast series Age of Miracles that really drew me in. Those were the ‘early’ days of ChatGPT’s meteoric rise and, although AI was gaining consumer attention, the arms race of hyperscalers hadn’t really commenced yet. Global policy was just starting to shift back toward nuclear. Proponents were optimistic about the regulatory environment under the new administration as momentum began building, uranium prices started rising, and more startups were launching.

But 2024 was really a watershed year for the industry as global adoption accelerated:

And it feels like we’ve hit a domestic inflection point in 2025:

So, we seem to have identified our next tick up the Kardashev scale as fueled by nuclear technologies. And with the vibe shift in full swing, investors have poured capital up and down the stack, with small modular reactor (SMR) companies alone receiving over $1.5B in funding in 2024:

In the near-term, we should see next generation reactors (SMR pods like Aalo’s) coming online in the late 2020’s. Regulatory loosening is certainly speeding deployment, and I could potentially see a world where this is accelerated further if power demands continue to go exponential. You’d think those AI gigabrains would find some power efficiency breakthroughs as costs become prohibitive, but we shall see. Either way, SMRs only fill the gap in a few years. Forecasts for viable fusion are (optimistically) targeting the early 2030s, and even that likely wouldn’t be scaled commercial viability.

Plus, as I referenced above, a true Type I civilization ultimately would require vastly more power than means like terrestrial nuclear will even be able to provide, assuming constraints around total uranium availability etc. This eventuality has led some founders to wonder how we squeeze more juice out of the only place in our solar system that’s solved fusion already… the Sun.

Ra is absolutely cookin’. But beyond our current solar arrays, how do you capture more of that sauce?

Two words: Space Lasers.

Aetherflux is easily the craziest startup I’ve found recently. Founded by Baiju Bhatt (co-founder of Robinhood), the team is launching an array of satellites into Low Earth Orbit (LEO) in order to gather uninhibited solar power and beam it back to earth. You heard that right.

“The idea is crazy. Crazy enough that it just might work. And the repercussions of such a system would be profound. Given the massive costs of terrestrial power generation, and the significant losses in power transportation, this scifi idea could upend power availability with a multitude of knock-on effects– everything from supply chains for the military to urban planning and nationstate infrastructure.

There’s precedent for this kind of technology with Caltech’s 2023 breakthrough successfully beaming 200 milliwatts down from space and DARPA’s recent 800 watt wireless power delivery over five miles. Granted, these were miniscule power levels: the DARPA example transported about enough energy to pop a bowl of popcorn. Both are very early, and clearly the Aetherflux team will have to harness significantly more power over vastly farther distances, but these examples serve as proof points that the idea can be done.” - Some Scallywags You Might Love (Words with Wynn)

Now, if they are able to get this technology working at scale, you could envision a future society that’s leveraging orders of magnitude more solar power than we’re snagging today. And if you really get the tech working… Well, I played enough video games growing up to know exactly where this ends up going:

Of course, Aetherflux’s vision is scifi sounding, but there are also startups building stepstones toward this idea. Lux Capital backed Reflex Orbital is actually deploying a fleet of satellites with mirrors to provide sunlight on demand. This kind of reflection could enable 24/7 farming or perpetual solar power. Ultimately, the cleanest source of energy we have is the sun, but it’s a matter of capturing its production in the most cost effective, least pollutive, highest efficiency way.

Mastering energy provides a clear throughline to measure the advancement of our species. There’s a reason that the myth of knowledge is Prometheus grabbing fire. From the first caveman’s embers to Edison’s lightbulb to the splitting of the atom, the capture and conversion of energy in all forms has defined humanity’s march forward. The Kardashev Scale merely provided a vocabulary for its measurement.

In our age of acceleration, the Kardashev Scale is as good a goalpost as any, though its climb should not be pursued with reckless abandon. Command of the stars is only possible for a society that lasts long enough to get there. Our ability to split the atom and harness its power brought humanity into “technological adolescence”, as Carl Sagan defined it. We achieved a level of energetic mastery of the magnitude that could be our undoing (see: nuclear armageddon). The next leg up that Artificial Intelligence represents feels very much the same. First we could burn down villages, then we could flatten cities, next we may be able to displace humans altogether.

To reach the stars will require a dance along a knife’s edge- pushing forward with all due haste while avoiding utter self-destruction. But frankly, I think that galactic samba is what will define any civilization. There’s a reason we speculate about Great Filters, and whether or not we’ve achieved escape velocity.

So yeah, the Kardashev Scale is but another means of measuring this race and looks like a pretty good goalpost to shoot for. But also, your neighborhood tech bro can probably chill the hell out about invoking it with religious reverence.

- 🍋

Subscribe now

I’m beginning to suspect that the art of accelerators is really just aura farming. Everything is sales, and venture is no different. You peel back the onion in incubator land and begin to see through the matrix. Sure, there’s a bit of intellectual butt-sniffing (founders pitching their vision, investors showcasing their ✨value add ✨), but the metagame in a world awash with VC dollars ultimately comes down to selling a perfectly substitutable product: cold hard cash.

This is why branding is key: perception is paramount. The venture profession is a funny kind of magic trick- finding the most brilliant people you can, and convincing them that your dollars have more value than those of your competitors.

That’s the matrix of all these incubators. There is no spoon, just a collection of strategies for getting on more cap tables earlier and culling thereafter.

The name of the game for many accelerators is to cast the broadest net possible, last long enough for statistics to do their thing, and then scream to high heavens about your early wins. Spray, pray, and yap away.

Investing in startups is a treacherous business, with only 1 out of every 3 making it to Series A, let alone IPO:

So, the earlier you go, the riskier it becomes, and the broader a net you should cast. This math is precisely why every accelerator works through much higher volumes (dozens of companies a year) than later stage investors that may pick about 15-20 companies per fund. In order for the Power Law to work its magic, there’s a sweet spot of how many bets you should place to optimize your odds of outperforming. If you want a gory analysis of the math, check out Jerry Neumann’s excellent post: Power Laws in Venture Portfolio Construction.

But in order to help skew the odds in your favor, regardless of portfolio size, there are three major components of the funnel to build:

• Source better companies (improve the average quality of your dealflow)

• Give your founders unfair advantages (support your bets better than the competition)

• Double down on your winners (have a robust means of following on later)

The best firms do some mix of all three exceptionally well, but here we’ll discuss how different players construct the top of that funnel. This is the dealflow engine that VC’s use to fuel their later stage funds, typically securing some kind of proprietary access in subsequent rounds. There are many ways to skin this cat, and I’ll only be highlighting a few specific examples. But if you’re a founder looking to really get a lay of the land, I’ve consolidated a mega sheet here listing over a dozen of the most prominent accelerators/incubators and how their workflows look.

As I’m sure you’ve noticed, I’m using accelerator and incubator interchangeably for the sake of this piece. A pedantic investor might bifurcate these models into incubators for the earliest of early (pre-MVP, pre-revenue) and accelerators as a means of speeding growth (funding PMF, teaching GTM, supporting scaling). The way I’m thinking about them all here is as top of funnel for your later stage/follow-on funds to cherry pick the winners.

There are a few interesting permutations of early stage strategies in the pre-seed/seed arena, and I broadly classify them as Classic, Advisory (Sweat Equity), and Freemium.

Classic: Classic accelerators running a standard cohort-based support model.

Y Combinator - The creme de la creme of startup accelerators. YC is the darling of Silicon Valley, and known the world over for the caliber of its founders, funders, programming, and team, to say nothing of its elder statesman Paul Graham. YC runs four cohorts each year of around 125 companies (a recent shift from only two batches per year). Strategically, they’ve consolidated their might in Silicon Valley, after running bi-coastal programming between Boston and California in the early days. They invest in companies through the YC Standard deal:

• 3-month program, now run 4x a year

• $125K SAFE for 7% ($1.8M post) + $375K uncapped MFN SAFE

Techstars - Techstars runs a cohort-based model similar to YC with a slightly different strategy for scaling. Techstars took a more national approach, partnering with cities and corporates to spin up localized programs, each run by a sector-specific managing director. In many ways, Techstars was both beneficiary of and contributor to the U.S. ‘technification’ wave as Silicon Valley began to dominate the U.S. economy and cities across the country pined to energize their own startup ecosystems. This was a net positive, but strategically created more of a federation of Techstars franchises than a monolithic brand like YC. They run over 50 accelerator programs across 30 locations, backing 700-800 companies a year. Ultimately, this decentralization has allowed Techstars to lean into city-specific superpowers (ex: fintech in NYC) while producing volume, building national brand aura, and securing pro rata across the board. The challenge though has been quality control, with founder experience relying more heavily on the caliber of the in-market manager. Techstars updated their investment terms a couple years ago, moving toward YC’s standard:

• 3-month accelerator program, with cohort cadence differing by market

• $20K for 5% Convertible Equity Agreement + $200K uncapped MFN SAFE

Boost – Boost is not an accelerator, but a high-volume pre-seed fund with a number of accelerator-adjacent mechanisms for capturing particularly early deal flow. Adam Draper and Co. have built a stellar reputation as first check investors, moving with conviction and speed. The fund’s website says that they average one deal per week, investing $500K checks. On top of their traditional dealflow though, the Boost team capitalizes on specific thesis or extra early opportunities with a few other programs. They’ve created a Founders Start equity grant as well as topic-specific cohorts like their recent Bio Residency or pitch competitions with partner funds (s/o Arian and Earthling VC):

• Residencies: $500K for 10% ($5M post); ~4 weeks

• Pitch Competition: $100K SAFE for 4% ($2.5M post)

• Founders Start Fellowship: $50K for ~3% ($1.5M post)

• Fund checks $500K (val presumably TBD)

Advisory (Sweat Equity): Services and support in exchange for a nibble at the apple.

Capital Factory - Capital Factory is the biggest, baddest ecosystem builder in the state of Texas. The Austin based firm runs a sprawling collection of businesses including major conferences, coworking, founder education, corporate scouting, ecosystem building, a DARPA Commercial Accelerator, and a family of funds. The team has evolved a unique model for supporting founders at scale through its All Access program, which engages around 100 companies a year. All Access is more akin to bringing an institutional advisor onto the cap table than a formalized 13-week accelerator program. What this means in practice is that startups exchange $100K worth of equity (capped at 1% ownership) for access to a suite of services to support their scaling. This includes everything from Capital Factory’s dedicated investor relations team through its corporate partnerships and Defense Innovation Center. It’s a lifelong relationship where founders can tap the various components of Capital Factory’s platform as their challenges evolve. This structure and ecosystem approach also allow Capital Factory to engage companies across the lifecycle of their business. The capped $100K equity grant is less dilutive for later stage companies and drives a broader collection of founders within the ecosystem orbit. This, in turn, helps to engage a deeper set of partners- corporates, defense partners, multi-stage investors. Intrinsic to the agreement is a guaranteed allocation in the startup’s next priced round, which helps to feed Capital Factory’s follow-on vehicles.

• Lifelong relationship; year-round, ongoing support centered around cornerstone events

• $100K advisory equity granted, capped at 1% ownership

• Fund checks ranging from $100-$250K

Freemium: Offering resources for founders, as a means of getting them in the door.

Plug and Play - Plug and Play is a bit like an external innovation arm for corporate partners, acting as a sounding board for their biggest problems and then sourcing startups to solve them at scale. What originally started in the Lucky Building in San Francisco (early office space of Google and PayPal), has now expanded globally to 125 accelerators working with over 2,500 startups each year. What differentiates Plug and Play is its equity-free accelerator offering. The firm has aggregated over 500 global corporates for whom it scouts startups; these partnerships subsidize the cost of Plug and Play’s programming. This arrangement is mutually beneficial as it offers founders close access to corporate partners, which may result in customer relationships, while simultaneously giving Plug and Play a privileged look into the market’s appetite for various startup solutions. The team is then able to accelerate traction achieved through their matchmaking with investment from their fund thereafter (investing in over 200 startups a year). Additionally, as opposed to others listed here which focus pre-seed/seed, both Plug and Play’s accelerator and its fund are stage agnostic, which aligns with the varying scale and need of the corporates it serves.

• 3-month equity-free accelerator program; lighter touch than YC or Techstars

• Fund checks vary, depending on size, round, etc.

Launch - Launch is Jason Calacanis’s brainchild. The platform is a multi-layer founder funneling media machine, built on the back of Jason’s brand and reach. Launch supports founders across three segments of ‘early stage’ via its Founder University, LAUNCH Accelerator, and Fund/Syndicates. Founder University is the top of funnel 10-week pre-accelerator helping the earliest founders with the basics (MVP, business model, marketing, pitch), and runs at the highest volume with 3-4 cohorts of 80+ founders per year. Typically, the team then invests in the top 10% of Founder University graduates and/or cycles them into the LAUNCH Accelerator, which is more concentrated on smaller batch companies starting to get real traction. LAUNCH Accelerator whittles it down significantly to 10-14 founders per cohort a couple times a year. Jason and team then follow on in the breakouts via his fund or syndicates. The funnel is broadened at the top by Jason’s massive media presence (All In, This Week in Startups, and Angel) that can also serve as microphone/marketing tool for the Launch portfolio companies as well.

• Founders University: $25K for 2.5% or $125K for 7% ($1M or $1.8M post, respectively)

  • Tuition is $500 for founders, reimbursable if you attend all programming

• LAUNCH Accelerator: $125K for 7% ($1.8M post)

• Fund/Syndicate Checks: variable

Note: take all these check amounts, cohort cadences, and class sizes with a grain of salt, as all of these various programs are constantly evolving their models.

Regardless of model, starting an accelerator is a herculean lift. You can see from the examples above how many are financed through accompanying businesses, whether that be media, events, corporate partnerships, or some mix. It’s a very heavy flywheel to start. You need your programming to have a distinctive brand, your ecosystem to offer unique advantages, and your model to survive long enough for these bets to pay off. To run more intensive services requires a variety of resources, none of which are cheap, particularly if you’re going to have a physical presence. The management fees on a pre-seed fund alone likely don’t cover the requisite burn either. So, you basically have a deeper J-curve than traditional VC funds sitting on top of an inferior cost structure. This is compounded when you consider that most accelerators are engaging startups at the earliest of stages, thus have the longest time to liquidity even if their investments do well.

You also need a unique wedge to overcome the adverse selection of being a nascent venture platform. Multiple factors cause this inertia. Any founder worth their salt, especially with previous successes, is going to gravitate toward established brands with track records of excellence. Consequently, your talent pool could potentially be comprised of less experienced founders or less compelling ideas. You’re also likely to overindex on first time founders, as they are less likely to have the pitch polish and investor network to win funding on their own vs. repeat founders that have previous touchpoints with investors as well as other builders (even if their first rodeo wasn’t a slam dunk success). This adverse selection alone could be reason enough for broadening your portfolio bets, in hopes of catching outliers in an inferior pool of options.

As with everything in venture, there’s also a massive power law at play. The YC’s of the world have created enough density to form a supermassive black hole for talent. A select few players have the kind of mindshare that becomes a self fulfilling prophecy- talent attracts talent creates winners attracts talent… Thus, new accelerators often draft off the success of a bigger brand (ex: a16z Speedrun) or counterposition through a very specific wedge. A prime example of this would be something like Harpoon’s Black Flag program, which I wrote about last week. Harpoon has established itself as one of the premier defensetech investors, and has leveraged this moat to launch a national security-centric accelerator program called Black Flag. They’ve also augmented the program’s gravitas by partnering with other heavy hitters in the space like Shield Capital and In-Q-Tel (the intelligence community’s VC arm). In all cases, it still requires selling the best builders on why you’re the partner of choice, and why they should pick you over other S-tier funds.

Hence, aura farming. You’re selling the dream that “Uber once passed through these hollowed halls! And you could too!” to fuel the self-fulfilling prophecy you need to perpetuate. Like the startups they hope to back, new accelerators must build the magnetism of a messianic cult. Whether that brand’s whacky, exclusive, aloof, patriotic, or ‘mad scientist’, working with founders is a human business and you’re cultivating a tribe around your banner. Broader reach and a more polarizing presence expands the pool of possible founders for matriculation while concurrently filtering for the builder archetypes you want to back.

That’s an element of why so many of these platforms have media-savvy partners at their head. Adam Draper, Joshua Baer, Jason Calacanis, Garry Tan all have commanding media footprints. Garry Tan has been particularly adept in this regard, scaling the YC media machine while simultaneously leading from the front with his own massive Youtube channel and Twitter following. The idea of VC as media is not new. We’ve seen this strategy proliferate as the industry got more competitive, and players relied more heavily on brand to differentiate. There have been acquihires like Turpentine (a16z), DIY’s like Pirate Wires (Founders Fund), and any number of other independent players like Sourcery, TBPN, and Not Boring that have leveraged mindshare into investment access.

But, with all this smoke and mirrors, founders should be mindful of the statistics at play when considering the value of accelerators, whose capital often isn’t cheap. Fellow founder referrals are gold in this business, and that can help to see past the patina online. “Sure, you have two unicorns in the last decade. Out of how many startups backed? A billion? Sick, your success rate is actually worse than random chance. Oh, and you gave my buddy a check and were nothing but extractive thereafter? No thanks.”

This is one reason I have so much respect for Garry Tan sharing YC’s hit rate:

But even this billboard includes a bit of subtle misdirection – conveniently picked 2018-2020 cohorts (a period which would have been primarily software startups massively marked up through COVID + ZIRP thereafter) vs. the total life of YC funds (2005). It also represents TVPI, which isn’t necessarily realized gains. This exemplifies yet another one of the funny dynamics of VC land: no one is incentivized to show their scorecard.

Again- spray, pray, and yap away.

It’s even funnier when you begin to learn how many VC’s sing the praises of deals where they got absolutely hosed. Restructuring, pref stack, dilution, or any number of other factors could mean that an early Reddit investment was actually a total wash, yet investors have every incentive to absolutely celebrate the founder in public and champion that position as their genius investing acumen.

Of course, economic reality catches up sooner or later – DPI, DPI, DPI. But that’s typically shouldered by the begrudging LPs on the back end, unseen by the founders dropping into the top of funnel. You can run a shit accelerator for a decade before anyone gets wise to the realities of the situation. And if you’ve got a cash flowing side business to boot, you can literally run an accelerator long enough to all but ensure that you eventually have a winner. The law of large numbers would suggest that, unless you’re utterly incompetent and your dealflow is trash, you should be able to get into at least a couple unicorns if you’re ripping hundreds of angel checks a year on a long enough time horizon.

Conclusion:

I got a bit cynical here. I really should clarify.

I don’t mean to say that the praying and spraying is all bad, although it can carry that connotation. Realistically, you kind of have to play a very large numbers game if you’re investing at the earliest stages. Data is limited, markets are unclear, and mettle is unproven. The accelerators that are run very well are a legitimate boon to the innovation ecosystem. They empower early founders with resources and expertise they may otherwise lack, and help to professionalize many, many ‘two guys and a laptop’ pipedreams into formidable businesses.

But founders should be thoughtful about just how much is marketing in all these lofty promises. DYOR, NFA, find referrals.

For the other side of the coin, young investors should also remember to be discerning as they see those “XYZ-backed” slides in decks they review. These accelerators work at such volume that their stamp of approval can’t be taken as an outsized mark of confidence, particularly when they’re playing a game across hundreds of companies versus the dozen or so picks you might make in your more concentrated fund.

All to say, accelerators, incubators, and especially ✨ecosystem builders✨ are a very funny business. No one’s incentivized to share their scorecard, the human mind is very bad at grappling with power laws, and even the best will produce 80% duds. Caveat emptor.

- 🍋

So, when ambitious young bucks come knocking on your door, the unfortunate reality is that you have to start being more discerning about taking that “coffee chat to pick your brain!”

And that sucks. A paucity of time plagues every ambitious individual.

I am eternally indebted to each and every person that’s ever taken 15 minutes out of their god-given life to help shed some light on my path ahead. I am extremely cognizant of the many small ways in which mentors have helped my career. As such, I try to do the same.

But, perhaps you can see the tension.

It’s tough to take the 20th ‘So, how ‘bout them ventures?’ coffee chat from a sophomore that just watched Silicon Valley for the first time. It’s fine if the All-In pod piqued your interest in douchey VC’s– that doesn’t mean we should meet up quite yet.

If you’re just knocking on the door of startups/venture, I’ve got you covered. I’ve drafted a megadoc of resources (books, pods, blogs, etc.) for anyone trying to break into the scene. I called it Your MBA in Venture. It’s got more content than you have hours in the day, and can easily cover your first six months learning about startups. Hell, it’s probably got enough to help you ace your first interview.

Now, we’ve got you cookin’. You know what a SAFE is and you’re ready to learn more. Or maybe you’ve already got that Venture 101 knowledge out of the way– you’re part of your uni’s ventures club, you’ve worked at a student startup, or maybe you’re already in tech.

Even better. Hit me back with any of the following:

• A couple startups that interest you, and why?

• A couple startups you think would interest me, and why?

• A couple startups I’ve written about, and why I’m wrong?

• A couple startups in our portfolio, and why they make you question our judgement?

• What specifically about venture/startups, and why would I be a better resource than AI?

• How can my firm improve our moat? What’s our mousetrap missing?

• My most (recent) misguided Tweet and why?

• The coolest shit you’ve built recently.

• The new, shiny sector we’re not thinking about yet.

I want to pay it forward, but I need a filter.

I don’t want to meet the casuals, the tourists. I’m looking for the weird ones.

I eat, sleep, and breathe this business.

If you do too, let’s chat.

- 🍋

Subscribe now

I recently came across Mike Annunziata’s piece, Five Business Models We Love, and its reference to Morgan Housel’s Betting on Things That Never Change.

Rereading the latter was a good refresh for my investor psyche, and exploring the former a good lens for my own startup scouting. Perusing Mike’s piece was stimulating, but learning requires action So, I thought I’d apply his framework to the Black Flag 100 to see who emerged.

The Blag Flag 100

Black Flag is a program spun out of Harpoon Ventures, and is building one of the best deep tech ecosystems around. It’s run by Harpoon in collaboration with Shield Capital and In-Q-Tel (the U.S. intelligence community’s strategic venture arm), and aggregates a wealth of resources for founders building in national security including non-dilutive funding support, a deep bench of mentors, and direct connections with leaders across the government.

A side note I’ve been thinking about –

It’s interesting that they also emphasize branding + design as an element of their support:

It may sound soft for the hard tech space, but I’ve really doubled my conviction in the importance of branding (and by extension, distribution) for deep tech startups. Many are run by gigabrain engineering types building incredible things, but ignoring the battle for attention puts them at a distinct disadvantage. A sexy brand, and a well-distilled mission helps to attract customers, talent, and investment (see: Rainmaker as a marketing master class).

I believe that this is why we’ve seen the explosive trend of hyped launch videos coming out of the likes of YC, probably emerging on the back of Anduril’s badass product videos last year. And more specifically, I believe that branding and/or marketing reach is one of the unique ways that an external partner can really add value for founders. Why else do you think every VC has a podcast?

A great example of this, and a firm whom I respect for it, would be Space Cadet Ventures– all of their work is sleek, their decks incredible, and their LP drops differentiated.

The Black Flag 100 itself is an aggregation of startups nominated by 300+ late stage deep tech founders. Their total universe was 500+ companies, curated down to the top picks by the Black Flag team. A couple stat sheets on sector and stage:

So, they’ve got the bona fides for some interesting deep tech picks, and I thought the Black Flag 100 would be a ripe ground for filtering startups through Mike’s framework to see what caught my eye.

Five Business Models We (might) Love

1. Recurring Revenue Infrastructure

These are companies that build real-world systems and monetize them with usage-based or subscription pricing. The business grows as the network grows, and the revenue is predictable.

Aetherflux

Aetherflux is insanity. Founded by Baiju Bhatt (co-founder of Robinhood), the team is launching an array of satellites into Low Earth Orbit (LEO) in order to gather uninhibited solar power and beam it back to earth. What?

The idea is crazy. Crazy enough that it just might work. And the repercussions of such a system would be profound. Given the massive costs of terrestrial power generation, and the significant losses in power transportation, this scifi idea could upend power availability with a multitude of knock-on effects– everything from supply chains for the military to urban planning and nationstate infrastructure.

There’s precedent for this kind of technology with Caltech’s 2023 breakthrough successfully beaming 200 milliwatts down from space and DARPA’s recent 800 watt wireless power delivery over five miles. Granted, these were miniscule power levels: the DARPA example transported about enough energy to pop a bowl of popcorn. Both are very early, and clearly the Aetherflux team will have to harness significantly more power over vastly farther distances, but these examples serve as proof points that the idea can be done.

I view this project as one of those unique startups building at the intersection of multiple declining cost curves while capitalizing on the second-order effects unlocked by recent technological breakthroughs: collapsing launch costs, advancements in sensors/transmitters, on-board compute, solar panel efficiency, emerging space economy supply chains, etc.

Hence, it’s a long and ambitious bet, not unlike the SpaceX and Vardas of the world, but the spoils of success would be massive.

This fits well within Mike’s framework, with a growing network value accruing as the underlying infrastructure develops, which will presumably be highly sticky given the insane capital cost of entry in combination with the technological moat you will establish should you be able to make the technology viable.

2. Full Stack Vendors in High Trust Markets

In regulated or risk-sensitive environments, customers prefer one responsible counterparty. If you can deliver a complete solution—hardware, software, service—you build trust and stickiness.

Nuclear

I don’t want to beat a dead horse on nuclear, but I think the entire category pretty well fits this idea. Multiple startups in the emerging nuclear renaissance were included in the Black Flag 100, including hometown (Austin) heroes Aalo Atomics and Hexium:

It’s a developing race to market for SMRs with the Department of Energy recently announcing their selection of 11 reactor companies to help expedite the testing of advanced designs. But more than that, we’ve seen a resurgence across the entire value chain from startups like Founders Fund backed General Matter (enrichment) or Standard Nuclear (fuel production) through to ‘AI for Nuclear’ software like Atomic Canyon. TBD precisely where value accrues here, but it does seem like there’s significantly more competition at the SMR/reactor link of the chain vs. even more regulated components like enrichment where the government and commercial players will likely source from only one or two vendors. That said, the mind-boggling power demand of the AI arms race could represent a multi-winner market on the SMR side.

Andromeda Surgical

Maybe alittle on the nose for Full Stack Vendors in High Trust Markets, but Andromeda is racing ahead with an autonomous surgical assistant, starting with endourology. They’ve built a tablet-operated platform to help urologists perform Holmium Laser Enucleation of the Prostate (HoLEP) – a gold standard but technically demanding procedure for enlarged prostates.

The integration of robotics and automation into the operating room continues to be slowgoing, despite many of the biggest players in the world attacking the opportunity. Intuitive Surgical has been working on its Da Vinci platform for over two decades, initially operating as a laparoscopic surgical assistant and now layering in incremental automation.

Andromeda is intriguing in that it’s already semi-autonomous, with the surgeon today initiating and supervising the procedure. Where systems like the Da Vinci rely on joystick-based mirroring of the surgeon’s fine motor movements, the Andromeda system is more akin to auto-pilot that commercial airline pilots utilize today– manual for a few touches, but largely self-directed with supervision. The startup has been accruing validation for its technology, performing over a dozen clinical procedures in Chile and New Zealand.

The first mover advantage within robotic medicine could provide a significant moat, as well as a well-fortified beachhead for product expansion. Once trust has been built at the hospital buyer level, it is likely that adding additional use cases to your existing system and selling those through will be easier than a de novo device or player attempting to break in.

3. Tools That Become Platforms

Start with a narrowly useful product. Over time, expand into adjacent workflows, integrations, or systems of record. These businesses often start as tools and evolve into infrastructure.

The ‘Tools Become the Platform’ lens pairs nicely with another one of my favorite deep tech frameworks: Packy McCormick’s Vertical Integrators thesis.

Packy’s two-part treatise is summarized above, and I think it is a useful mental model for these kinds of hardware heavy industrial disruptors vs. vertical/horizontal software type products that require a different approach for consolidating their value chain.

Durin

Durin could be a Gundo Bro poster child.

The startup is helmed by the fearless Teddy Feldmann, with grand ambitions of building a next generation integrated mining titan to secure American minerals. The stakes are huge, the tech is dirty, and the market is massive.

Ambitious from inception, Durin calls its shot in both name and vision. The startup’s name pays homage to the altar of Thiel, pulling directly from the Lord of the Rings universe to invoke mythical lore and Silicon Valley progeny. Tolkien’s Durin was the greatest of the dwarven kings, and referencing Middle-earth mythos is a beloved tradition of American Dynamist darlings:

• Palantir (data analytics) - named for the seeing-stones that allowed distant communication and surveillance across Middle-earth

• Anduril (defense tech) - Aragorn’s reforged sword, the Flame of the West

• Erebor (crypto bank) - the Dwarves’ ancient stronghold and treasure hoard

The company itself is starting with a specific tool in a specific link of the mining value chain as their first product to gain a foothold in the market.

For context, developing a new mine is an extremely long and complex process, with a number of players shouldering different levels of risk all across the lifecycle:

Typically, a mine starts with land acquisition and prospecting rights (usually by ‘wildcatters’), followed by geological surveys and surface sampling. Once there’s enough data to form a hunch about a particular parcel of land, then comes core drilling where the wildcatter will hire a third party contractor to drill and pull dozens of deep samples to validate the underlying minerals. These samples are then fed back to geologists, labs, and auditors to thoroughly assess the value and begin making the case for developing the underlying. Given the incredible investment required to develop a full-blown mine, you can imagine the importance and thoroughness of this validation step (many cores to be drilled). Once a compelling case can be made for development, the wildcatter or junior miner typically sells the land to or partners with major mining players with the scale, balance sheet, and expertise to build the mine itself.

Exploration and core drilling are the steps that Durin hopes to improve, finding and validating more mineral deposits faster to accelerate the development of domestic and allied rare earth mining operations. The team is integrating a number of off the shelf parts to build semi-autonomous core sampling rigs in order to perform this piece of the value chain cheaper and faster.

Strategically, it makes sense that a startup would attack this link. It represents more consistent work with less risk and lower capital requirements. But even within core drilling, Durin’s hoping to build an edge and operate with improved margins by developing their own rig and sensor suite as opposed to existing players that are more service contractors (buying/leasing rigs from OEMs and providing the labor to operate them). By their estimate, core drilling is a highly manual endeavor and can be as much as 80% of the costs of an exploration project. Teddy & Co. believe they can provide this service in a more streamlined offering by manufacturing a semi-autonomous drill in-house and operating it more efficiently.

From there, Durin can evolve from tooling to platform, layering assays, geophysics, and analytics into a vertically integrated exploration engine.

4. Data-Exhaust Flywheels

Some companies generate valuable data as a natural byproduct of their core operations. That data can be used to build intelligence layers, drive new products, or create entirely new revenue lines.

Orchard Robotics

Orchard Robotics is building an agricultural data machine. If data is oil in the age of AI, then this team is prospecting the Permian. Where many startups have stumbled building expensive robotic labor solutions for the agricultural sector, Orchard has taken a different tact.

Charles Wu and team have developed a mountable camera system with the ability to capture over 100 images per second, recording data about every fruit on every tree within a grower’s orchard. For scope, orchards can have millions of trees growing hundreds of millions of fruit across thousands of acres. It’s an incredibly complex data question to provide tree-by-tree analytics in an economical way (farmers historically leaned on statistical sampling to make harvest-level decisions), especially given the terabytes of data required in generating reams of images for every tree. Couple this with the challenge of developing farm-sturdy hardware that can operate rain or shine, and you’ve got a particularly thorny problem.

The fidelity of this data is key in driving effective decision-making processes for farmers, and can be gathered in a non-intrusive way by simply strapping Orchard’s camera to existing equipment like tractors and carts.

Orchard then offers its FruitScope OS as a comprehensive farm management software, leveraging the data provided by its camera stack. This razor+razorblade business model, placing low margin hardware coupled with a high margin software stack, is interesting because it provides Orchard with a unique strategic position. As opposed to other startups that must scramble to gather data for their capital-intensive robots, Orchard has the unique vantage of picking from commoditized hardware players in the future and operating as a higher margin intelligence or data layer for its partners.

This appears to be where the company is potentially going with its forthcoming Canary product, which it is marketing as an AI powered decision-making tool to improve the entire growing lifecycle - pruning, spraying, irrigating, hiring, harvesting, marketing and selling, distribution, etc.

5. Real Asset Arbitrage

These businesses find overlooked physical systems—cooling, sensing, RF, etc.—and apply modern technology to reprice risk, lower cost, or create new economic leverage. They often don’t look exciting from the outside, but they work.

Parallel Systems

Parallel Systems is a bit closer to home. This one isn’t actually in the Black Flag 100, but fit too well to pass up. I live about 50 feet from the train tracks that run through downtown Austin, and the daily screeching steel of American Industrial Might led me down the wormhole of “Why isn’t there a SpaceX for rail?”

Founded by three early SpaceX engineers, Parallel systems is developing autonomous, battery-electric freight vehicles.

What interests me about Parallel Systems is not that they will disrupt traditional rail (they might do that), but that they found an orthogonal way to attack the industry that aligns their business model with the existing oligopoly of players.

Rail has been getting crushed by freight for the last decade for its lack of flexibility and service deliveries:

Parallel’s wedge into the market is enabling uneconomic (low density) lines to become viable, thus expanding the market for railroads and helping them to take share back from freight. And when you consider the difference in market sizes, trucking is 10x rail by revenue, even a fractional shift in share could be a massive outcome for Parallel.

Then they become infrastructure. Parallel is a heavy industry play that helps their customers win. It’s a B2B2B value chain whereby they support the Tier 1 railroads, allowing them to improve their service and expand their offerings to shippers. Shippers are happier with the consistency and additional optionality, and thus steer some of their choices from freight to rail. Once integrated, Parallel’s systems become an irreplaceable component of the rail service offering, and the regulatory challenges they’ve scaled are digging an increasingly deep moat in their wake.

The company also has an interesting second order effect of reducing carbon emissions and abating overall congestion. They’re utilizing an electric drivetrain for their system, which saves of course on pollution at the margin, but where I find their model compelling (from an externalities perspective) is that success for Parallel potentially reduces trucks on the road and city-wide congestion around major ports and rail hubs.

Bonus: Airship Industries

A Black Flag 100 that does fulfill a similar idea to Parallel is Airship Industries. Backed by Packy and pointed to as an example of the Vertical Integrators that he loves, you can learn more here to decide whether or not founder Jim Coutre is full of hot air.

Conclusion

Perusing the pirates of the Black Flag 100 fires me up for the future:

It’s a treasure map of builders and businesses trying to upend industries. Mike’s framework is an excellent tool to sharpen one’s thinking about which of these players may emerge victorious. Of course, compelling business models alone do not unicorns make (it’s the blood, sweat, tears, and grit of the founders pushing them forward), but studying how they interact and where they sit can help inform your understanding of why some innovations win.

The privilege of being a venture capitalist is meeting the ambitious minds ideating these technologies and watching them reinvent the business models they enable. It was a pleasant surprise to see a few of our founders grace the Black Flag 100, and even more exciting to note the new ones I’d like to meet.

Finally, if you’re a pirate that may have missed the list. Even better. Drop me a line.

- 🍋

Subscribe now

Last month, Anduril and Meta announced a partnership to help push the boundaries of augmented reality (XR) for the U.S. military. The details are fairly high level at this point, but the gist is that Anduril will be the prime with Meta working as a technology provider. It sounds like Anduril will be responsible for manufacturing the hardware itself (helmets, goggles, HUDs etc.) and integrating the tech stack (Eagle Eye) with its broader Lattice network as well as other DoD technologies, while Meta will provide its XR expertise, IP, and AI models. This is an incredibly exciting ‘getting the band back together’ moment for Luckey and Zuckerberg to reunite years after Palmer’s unceremonious removal from Meta in 2017. In his Core Memory interview (linked below), Palmer has said that this collaboration allows him to re-access much of what he developed internally while building Oculus then Meta, and supercharge it with their work at Anduril. The announcement builds on the momentum of Anduril’s previous XR coup back in February: taking the prime position on the DoD’s $22B Integrated Visual Augmentation System (IVAS) program from Microsoft.

This got me jazzed, not only because it portends a future of technomancer American supersoldiers..

…but because I peeled the onion back a bit, and went deep on Palmer Luckey’s interview with Ashlee Vance. That discussion gets into the meat and potatoes of what this partnership means and what Palmer really has in mind for the future of reality-enhanced American warfighters.

In that meandering hour, there was a specific minute that really made my ears perk up:

“So, the way that I look at something like Eagle Eye is that it should not be one display system that does everything for everybody. It’s actually an architecture that allows- like, EagleEye can basically, it can host many different types of vision augmentation systems. It can host head mounted displays from multiple different companies, including companies that don’t even work directly with Anduril. [We’re] building an open architecture where the government owns the interconnects. They own the standards and they can work with other companies to build different displays that plug in.

The reason for that is that I don’t think we can build one display that does everything for everybody, right? The guy who is a logistician needs a different thing than the guy who's a helicopter pilot who needs something different than the guy who's an infantryman breaking down doors in the middle of the night. And in terms of the units, which is what you're asking about, I think that within the decade, you're going to see a head-mounted display of some kind on basically every single head in the DoD.

They're going to use it for different things. Some guys are just going to be checking their emails on a secure terminal. Others are going to be shooting people in the face. And you need to be able to support all those use cases. I don't see this as just like 10,000 units or 20,000 units. I see this as outfitting the entire armed forces. And so everything I'm doing is to try and build an architecture that scales to that.

…

A guy who's like, let's say, an army aviation maintainer who's working on helicopters, he probably needs something that is like a lightweight pair of Oakleys that is able to display work instructions, remote help, labeling particular things, checklists. He does not need an ultra-wide field of view, thermal, UV, hyperspectral vision augmentation system like a frontline infantryman might. And so you kind of have to build something that unifies those two things so they can share one set of software, one set of apps, one set of common standards.” - Palmer Luckey on Core Memory

Anduril isn’t building a product, it’s launching a platform. And this framing sounded awfully familiar to me:

“I'm pretty sure glasses and holographic presence and AR is going to be a completely ubiquitous product, right, it's just like everyone who had a phone before replaced it with a smartphone and then a lot of more people got smartphones. If all we get is all the people in the world who already have glasses upgrading to glasses that have AI in them, then this is already going to be one of the most successful products in the history of the world, and I think it's going to go a lot further than that. So I know there's that.

There is the thing about controlling our own destiny. It's strategically valuable. You know we did this calculation or estimate at some point where it's like how much money do we lose from our core family of apps to the various taxes that the platforms [Apple’s App Store, Google, etc.] have to like, when they tell us we can't run the ad business the way that we think we should be able to.

When they tell us we can't ship certain products, that way people use the things less or like them less, and it's hard to exactly estimate it. But I think we might be twice as profitable if we own the platform or something. I think from that perspective that's worth a lot, just from a pure dollars perspective, which is not primarily how I come at this stuff…” - Mark Zuckerberg on Acquired

The latter Zuckerberg quote was the inspiration for another Words with Wynn, Pieces on the Board: The Power of Platforms, where I basically pointed out that emerging AI startups and legacy corporates (see: OpenAI investing in Figure, Google investing in Apptronik) were all too happy to pour mountains of treasure into the coffers of humanoid robotics companies for a call option on a platform that could dwarf anything we’ve ever seen. Should that become the case, no one wants to be beholden to a toll-extracting App Store equivalent.

“Owning your own destiny [platform] has proven time and again to be fundamental in cementing network effects with hardware and user lock-in within your ecosystem. It’s tech strategy 101.” - Me, Pieces on the Board

Thus has been the logic with Zuckerberg’s massive VR investments as well.

Spelled out by the man himself, Meta’s unbelievable R&D bill potentially makes sense if an augmented reality piece of consumer hardware stands to be the next iPhone:

“AR/VR was all the hype around the time that Zuckerberg rebranded Facebook to Meta, and Wall Street torched him in the ensuing years for the company’s massive investments into augmented reality technology as the sector fell out of favor. But when you frame it within the context of owning the next platform, their hefty R&D bill starts to make a bit more sense. You can basically run a loose NPV for what that’s worth against the cash sink to develop the technology today. This theoretical number could be so huge that it’s rational to continue your investment as an option on a gargantuan future market. Whether or not AR/VR becomes the dominant platform is anyone’s guess, but if you’re the founder and controlling shareholder of Meta, and your timelines are measured in decades, there is a serious logic in betting on (and willing into existence) augmented reality if the probability of its ubiquity is high enough.” - Me, Pieces on the Board

Now, perhaps this is a case where the man with a hammer sees everything as a nail, but I don’t think I’m exaggerating when I highlight the underappreciated importance of this alliance in supercharging Anduril’s dominance as a future prime.

The partnership sounds benevolent when Palmer phrases it as the government owning the interconnects with an open architecture for other companies to build into the Eagle Eye ecosystem. But inherent to that dynamic is building atop Anduril’s platform, with value accruing accordingly. Even if it’s not a direct toll-taking relationship like the App Store, building the substrate for the military’s entire XR stack (third party or otherwise) to integrate into is an enviable strategic position to occupy.

It’s all platform. It’s all ecosystem. And if Palmer Luckey has his beautiful way, it will all be computer. Delicious, delicious Lattice-enabled computer.

This collaboration has the potential to be a win-win-win: leveling up the American warfighter, irreplaceably cementing Anduril as the DoD’s XR dealer of choice, and giving Meta another avenue to capitalize their ungodly R&D spend. Plus, if history is any guide, we’ve seen tools of war push the cutting edge for technologies (GPS, semiconductors, etc.) that eventually bleed into the consumer market, which still represents a significantly larger TAM than military-grade headsets. Meta recognizes this, and Luckey discussed this interchange directly in his AWE interview.

That said, there is also room for the reverse:

“It’s the Fairchild Semiconductor angle that is most interesting, however, specifically Bob Noyce’s resistence to getting too entangled with the U.S. government. His opposition was not ideological, but pragmatic. From Chris Miller in Chip War:

‘'The computers that guided the Apollo spacecraft and the Minuteman II missile provided the initial liftoff for America’s integrated circuit industry. By the mid-1960s, the U.S. military was deploying chips in weaponry of all types, from satellites to sonar, torpedoes to telemetry systems. Bob Noyce knew that military and space programs were crucial for Fairchild’s early success, admitting in 1965 that military and space applications would use “over 95% of the circuits produced this year.” But he always envisioned an even larger civilian market for his chips, though in the early 1960s no such market existed. He would have to create it, which meant keeping the military at arm’s length so that he — not the Pentagon — set Fairchild’s R&D priorities. Noyce declined most military research contracts, estimating that Fairchild never relied on the Defense Department for more than 4 percent of its R&D budget. “There are very few research directors anywhere in the world who are really adequate to the job” of assessing Fairchild’s work, Noyce explained confidently, “and they are not often career officers in the Army.’

Noyce’s gambit — most fully realized at Intel, the company he started after leaving Fairchild Semiconductor, in part because a venture-backed company offered far more upside — obviously paid off. The consumer market had massively more volume than the military, which enabled far larger R&D and capex budgets, which drove improvements in chips far more quickly than would have happened under the government’s direction. And, in the end, the Pentagon benefited: what undergirded the defense department’s successful modernization in the 1980’s was the wholesale adoption of computerized weapons systems that were based on technology that was built for the consumer market. - Ben Thompson, Anduril + Meta, Silicon Valley and the Pentagon, Meta’s Motivations

Where early government R&D helped to kickstart the U.S. semiconductor industry, it was ultimately the massive consumer market that provided the demand that powered the industry to track Moore’s Law. History doesn’t repeat itself, but it can rhyme. Meta can benefit from governmental use cases which will push the bleeding edge, while the budget to fund this kind of technology may ultimately be sourced from massive consumer demand.

A key aside – virtually all of Meta’s cutting-edge R&D up to this point has been financed off of its own balance sheet, and Palmer has explicitly said that Anduril will be moving forward with Eagle Eye irrespective of winning the $100M Army contract on which the companies jointly bid (per WSJ).

But what’s this mean for investors?

The inevitable lemming rush of VC’s leaping to back VR/XR projects with a DoD veneer need to recognize that, more likely than not, they will be funding apps, features, and add-ons for the mothership’s platform. Not that some major winners can’t emerge – plenty of billion dollar businesses were built on Apple’s App Store – but that they will be backing second movers that are beholden to a platform beholden to the government. Pricing will matter and value accrual must be considered.

And history would suggest as much. Anduril’s previous acquisitions have been one-off drone or weapons systems that could be integrated into its software-enabled hardware ecosystem; low-margin weapons platforms powered by high-margin Lattice. These individual roll-ups didn’t accrue value or command multiples like that of Anduril’s integrated platform:

“There are a lot of really interesting defense capabilities that are out there, particularly in hardware, that have kind of advanced through the prototyping phase– they’re capabilities that have been funded with research dollars, they’re platforms that are incredibly interesting and potentially valuable. They’re not necessarily highly integrated on the software, autonomy, artificial intelligence side of things, and make a lot of sense for us to go in and say ‘If we were to go in and make this part of the Anduril suite of capabilities, it would be more valuable than standing alone…’

These are not traditionally venture backed companies… And you know, we’re not buying them at Venture Capital software multiples, we’re buying them at Defense multiples… Now, the risk here is that if you see what Anduril’s doing– I think we’ve done eight or nine acquisitions to date– if you see what we’ve done in this space it would be easy to say like ‘Oh look, there all these very cool exit opportunities if you were to start a defense company.’ I just don’t think that’s true. If you’re a defense company out there raising at 20, 50, 100x on in-year revenues, like, we’re not a buyer of those companies.” - Trae Stephens on The Generalist

Investors should also consider the extent to which these types of VR/XR Eagle Eye add-ons will have broader consumer applications. The Anduril x Meta collaboration is for modular helmets/glasses manufactured by Anduril with open architectures for other vendors, not improvements and directly retrofittable upgrades for Meta’s Quest line.

Additionally, we could very well see augmented reality play out akin to modern computing. Original mainframe type computers were adopted by enterprises that had business use cases and could rationalize the high price tag against the efficiencies that would be gained across their organization. This familiarized a generation with computers, which then came down the cost curve simultaneously to now-tech savvy users seeing the benefits of this hardware for their everyday life. In the case of XR, the initial breakthrough may end up being military uses where immersive data visualizations (troop movements, autonomous system coordination, intel, etc.) are a superior form factor to current computer + monitor setups, with advancements in this domain feeding back out to consumer use cases like gaming or enterprise work. Of course, Meta has already pushed relentlessly to drive down the price tag on its Quests, so (finally) unlocking a ‘killer use case’ for VR should climb an accelerated adoption curve vs. the emergence of desktops.

The second interesting takeaway was Palmer’s discussion of margins and the evolution of military helmets themselves:

“And one of the things that Anduril’s doing is actually taking extremely low margin and making a very concerted effort to drive the price of the system down because… the army has, call it, $20B to buy headsets. If I price them at $100,000 each, then that means that there’s some number of headsets they can buy. If I price them at a tenth of that, they can buy 10x as many headsets. Which world do you think Palmer Luckey wants to exist? I want this on everybody.

So, I am doing my damnest to make this as low cost as possible, way beyond probably what is rational. I’m confident I could extract more money out of the DoD by playing the game, riding the margins as high as I can, taking advantage of the fact that what we have is so far ahead of every other competitor. I do not have it in my heart to do that.” - Palmer Luckey on Core Memory

I believe this is genuine and laudable. I too want our men and women in uniform to have world class technology at their disposal, and all the better if it can be done at a price that inflates our deficit marginally less. Then again, it isn’t (and was never going to be) the helmets that will drive Anduril’s profitability. It’s the software stack to power them. Any CEO should be happy to place hardware for a modest profit if they can capture high margin SaaS fees in perpetuity on the back-end, and even better if the switching costs are astronomical (once integrated) and your customer turns like a titanic.

In reimagining our army’s headware, he also made an interesting point that struck me about the various component costs:

“You can build some really interesting things that are far beyond what the United States has done so far on ballistic helmets. And I’d say, one of the reasons the U.S. hasn’t done these types of technologies in the past- It’s not because I’m this genius that’s smarter than everybody. It’s because it didn’t make sense when the helmet was this thing that’s supposed to cost a few hundred dollars. When a helmet was supposed to be this cheap thing, you put it on your head, it needs to work, but the whole system is pretty cheap. If you’re instead building this kind of very expensive thing where you’ve got sensors, computers, radios, displays, optics, speakers, microphones- all this stuff integrated. You now need to look at it as an integrated product. And when you look at it as an integrated product, you always want it to be lighter, right? So you start to ask, where is the best savings per dollar? The best weight savings per dollar.

If I wanted to pull a hundred grams out of this helmet, how can I do that the cheapest? Well, I’ll tell you what. It’s not pulling a hundred grams out of the electronics. It’s not pulling a hundred grams out of the optics. It’s not pulling a hundred grams out of the microphones. It turns out that the place to invest money to save weight is in the actual physical ballistic shell itself.

So, all the sudden it actually makes sense, in the context of an integrated system that, ya know, costs quite a bit, to say ‘you know what, my helmet skin is going to cost $2,000 and in doing so it’s going to weigh 2lb less than it would have weighed otherwise.’ And it starts to make sense.” - Palmer Luckey on Core Memory

This evolution could transform a soldier’s helmet from an elementary shell into a more ‘exquisite’ weapons system, as it’s outfitted with increasingly sophisticated widgets depending on the job. Loading down the helmet component of a soldier’s kit with increasingly sophisticated hardware begins to change the cost equation as you explore the possible steps to reduce weight distribution across the board. Given that the marginal cost to further miniaturize already cutting-edge electronics is massive, you actually turn to innovation in the shell itself, the largest weight component. But what this made me consider as well was the profitability profile for an exquisite helmet system vs. a singular shell with some modular tech attachments (night vision goggles, radio on the side, etc.). Is this a more profitable piece of equipment to manufacture if you’re capturing a blended margin with integrated electronics etc. instead of plain kevlar buckets?

It also begs the question: To what extent does a more complicated, integrated system fall prey to the ‘totaled car’ phenomenon of recent years where insurers more readily write off modern cars as totaled because the complexity and cost to fix vehicles loaded down with sensors etc. has made it more cost prohibitive to repair them?

I’m sure that bigger brains than my own are going deep on these considerations, and I’d imagine that many of the tech components that will be integrated into the helmet will A) be modular in nature and B) will displace cost/tech that would have been borne by a soldier in other parts of their kit anyway (backpack, belt, etc.). So, I’m sure it’s a complex calculus.

Ultimately, I think that this is an incredibly exciting partnership. The prospect of Silicon Valley leaning into supporting American deterrence (or whatever euphemism you want to use for American military might) in a time of escalating global tensions makes me optimistic for the future of our national security. And that’s a good thing. The bloody detail of business strategy within the less-than-free market dynamics of navigating the military industrial complex will continue to evolve, and spawn a generation of fascinating case studies.

The tailwinds for startups building in defense tech continue to mount as conflicts like Ukraine and Israel-Palestine have been rude awakenings for Western militaries. Government budgets have reacted accordingly as the imperative to modernize has become self-evident; the agility that Washington is attempting to reinject into the military contracting apparatus is just one of many examples. I have high conviction that first movers like Anduril and Saronic will be able to continue compounding their advantages (integrated ecosystems, economies of scale, etc.) for outsized returns, but the size of these markets and urgency of demand will certainly contain opportunities for ambitious builders and savvy investors.

- 🍋

I figured I should consolidate all my writing in one place.

So here are those varying pieces:

• A Toast to Toasts

• The Twitter Ouroboros

• Adverse Incentives & Junior VCs

• Spotif[AI] Between a Rock & a Hard Place

A Toast to Toasts - 1.17.25

Toasts aren't given enough anymore. Weddings mostly, sometimes birthdays, and once in a work party blue moon is a glass clinked and a pleasantry orated.

That's a shame. There's something special about a toast well wrought.

I'd argue that toasting is the real purpose for your high school public speaking class. Not the bullshit college powerpoints or myriad factoid vomits you will deliver your corporate overlords. The class failed you for the opportunities you'll have to warm a room and cheer your comrades. The perfect toast on a shimmering night can be romantic, it can be impactful.

I found myself toasting twice recently.

Ukrainian Dinner

The former was Ukrainian Dinner. My very dear friend has a hairbrained tradition-- Ukrainian Dinner. He's not Ukrainian. No one at the party was. No one at its founding was. But that's the theme. Between borscht and vodka, toasts are required of all who attend. Passed down from a friend of a friend, it's now an annual tradition, and it brings about toasts that only a half decade affair can.

As such, you can't come empty handed. Sure, a first timer can mumble through a tip to the host's good health. But a repeat offender appreciates the gravity. It's an occasion to be inventive in flattery of your friends; it's not often we publicly praise the people that enrich our lives. Everyone brings their 'A' game.

There's something magical about it. The room warm with laughs and familiar faces, a rotating cast around the core contingent. Between bouts of conversation, a chair will creak back and the next speech is at hand. It's a collective act of creativity that nurtures the wit we often neglect.

Beach Birthday

For my friends' fortieth birthday, they went all out. A trip to somewhere tropical with her best mates in tow. Amidst the preparations, her thoughtful husband asked the attendees to come with toasts in-hand, to be delivered throughout the course of the vacation. The results were tremendous.

Little riffs delivered from the heart. Some short. Some sweet. Some quick. Some long. People played with ideas and remixed the ask to their liking. Themes emerged over the first couple, and were played with by later toasters. Some made us laugh and some made us cry, with an authenticity that only small silly circles can create.

In our bland, flavorless world, it's a creative complement to the breaking of bread. I lament that these brief moments of human connection are so infrequent. And so I say, a toast to toasts, and adding a bit more spice to your life.

// 🍋

The Twitter Ouroboros - 1.15.25

"The internet introduces a world in which the importance and the difficulty of having principles is skyrocketed." - Ben Thompson

This quote comes from Ben's discussion breaking down Meta's recent content moderation changes. The internet pervades my life, and as I've leaned into social networks increasingly over the last twelve months, I've started to see the ouroboros cycle of self delusion that can emerge.

You read, you write, you chatter, you meme.

It's all fun and games as you banter back and forth. Yet... at some point you start to see just how narrow your feed becomes.

Twitter isn't as inherently recursive as a platform like TikTok, whose 'for you' lives to reinforce your slightest indulgences, yet it's very easy to find yourself within a homogeneous pool of thought. I meander through Crypto Twitter, then Tech Twitter, then American Dynamist central, and suddenly they've all got blurring views converging on groupthink.

The more you lean into a space like Twitter, contributing, conversing, and clashing with its inhabitants, the more you find yourself engulfed by its information stream. And realizing just how extremely that broad stream flows in only one direction. This is self-evident, but there are multiple layers at play: some amount of increased exposure (more consumption), some amount of increased engagement (more contribution), and some amount of psychological rationalization (more contemplation). The latter, I believe, is the more subtle.

We all view ourselves as being above the indoctrination of these platforms and their bedeviling algorithms, but therein lies the danger. Like your friend with a coke habit-- not perhaps outwardly acting as an addict, but always toting it along on any given outing and always at hand for their compulsive indulgence. It isn't a single Tweet that indoctrinates an addict, but the subtle slide of months engrossed in a singular point of view. The more you discuss and the more you contribute, the more likely your sense of investment is to rationalize and erode your original viewpoints in the slide toward your repetitive platform exposure.

Some of this is being dramatic. But it's important self-awareness.

So, this year I'm looking at my information diet, I'm reexamining my principles, and I'm trying to double down on refining my own voice.

This brain blast was prompted in part by Mario Gabrielle's resolution to cut his own social media exposure. I've paraphrased or bastardized a few of his ideas above, but I'll end with its relevance in a world of AI:

"Surrendering to that momentum by donating your attention is a step toward losing ownership of your mind. It is too easy to find yourself adopting opinions and preferences you haven’t thought through but that are part of the fabric of these networks.

It has always been useful to be inventive, but it is increasingly essential as AI continues to bolt knowledge work in great hungry gulps. If it has not happened already, very soon, our greatest advantage will simply be the ability to think as ourselves — to rely on our unique accumulated training data and apply our singular algorithm. The more your data and algorithm adhere to the dominant monoculture, the less you will have to offer.

As much as possible, I would like to ensure that I continue thinking like myself rather than running someone else’s program." - Mario Gabrielle, Monk Mode

It's about getting smarter with WynnGPT. Discerning training and careful curation for feeding my own wrinkled wetware a higher quality diet.

// 🍋

Note: Article.app pieces are time-boxed, stream of consciousness for the most part. Expect errors, warts, and fallacies.

Adverse Incentives & Junior VCs - 1.9.25

“Show me the incentives, I’ll show you the outcome.” - Charlie Munger

Adverse incentives abound in venture. From “Founder first!” VCs shackled by fiduciary duties to “Long term capital partners!” needing liquidity in 24mo, it can be a confounding headache for the uninformed.

I think this is particularly prevalent when founders meet with Junior VCs.

Analysts/Associates are often stuck between a rock and a hard place, and many toe the line of safety to the detriment of founders. “Too early!” or “Not right now!” are often coded for “We don’t think you’re the team” or “We think this is Tinder for Dogs”. Responses are often mushy pleasantries devoid of substance.

But what leads to this behavior?

A few things:

— Venture is a long term game, and the longer you’re here, the more surprises you see. Tinder for Dogs might pivot to the Dog Walking Unicorn. Every VC is balancing feedback with keeping the door open to future conversations…

— … and many founders do not like being told their baby is ugly, even if feedback could be productive. So, many Juniors fall victim to a flawed risk-reward calculus that fizzles down to “We’ll circle back!”

— Plus, many Juniors believe they have neither the bona fides (builder experience, track record) nor the platform (founder/partner/title) to speak authoritatively in a way that will find founders receptive. Khosla can torch your idea because he’s Khosla; the proof is in the pudding. Johnny Junior can torch his reputation with a bit of spicier feedback poorly communicated.

Counterintuitively though, in seeking safety, many Juniors do injury to themselves and their brands as non-committal, wish-washy, fake, or memetic (“How can I be helpful!”).

The best investors have brands of being honest and providing quick “No’s” with actionable advice. In fearing polarity, Juniors often relegate themselves to the milquetoast pack of mediocrity.

I’ve done this. I’m not proud of it.

As I’m in the space longer, I get more comfortable with my voice, my value add, and where I think I have room to impart something thoughtful and where I should really shut my dumb mouth. But it’s still a balancing act. We’re in a very human business.

As frustrating as it is, founders should be aware of the game they’re playing and the incentives across the table.

// 🍋

Spotif[AI] Between a Rock & a Hard Place - 1.3.25

Spotify has to be in such a strategically frustrating spot right now.

If they integrate UGC (AI-generated music) from subscribers natively, they'll likely get a revolt from the artists + labels.

So, they're stuck waiting for an AI music platform to start disrupting them, at which point they'll be forced to acquire it at an exorbitant premium once the market/zeitgeist attenuates to this type of content. By the time that it's possible (or even allowable) for Spotify to enter that race, it will (definitionally) be won.

In a perfect world, it seems logical to me that Spotify would leverage its huge distribution advantage and simply empower its deep user base to become creators. A kind of Roblox judo move that keeps the created content on their platform and embeds additional monetization opportunities (tokens for generation, improved sound design, new subscription tier).

The classic 'creative destruction', attempting to cannibalize one's core business in order to attack adjacencies and shimmy up the next j-curve.

Unfortunately, this seems to be impossible (or incredibly challenging, to say the least). Attempting this kind of product expansion would likely be armageddon for its core business as the company was threatened by myriad lawsuits, artist desertions, and copyright infringements (how was the AI trained? on what songs? etc.). It would slaughter the cash cow in a recursive fashion, destroying the viability of the AI product launch at the same time as the user-base's original value proposition collapsed.

I think that Spoitfy is stuck over a barrel. The legal inertia from their agreements with record labels (supplier power - Porter's 5 Forces) acts as a hard stop inhibiting Spotify from taking the necessary steps in order to truly leverage the AI revolution (at this point).

And we've seen this with the platform's lukewarm attempts to interweave AI so far:

• AI DJ: a marginal improvement over the legacy Radio feature

• AI Thumbnail Generation for Playlists: a nice to have, but minimally improving an audio-centric platform

• Generative Playlists: text-to-playlist (type mood, theme, etc.). Seems to overindex on your likes/current playlists. Eh.

But perhaps this is a bit myopic, disregarding:

• AI generated podcasts

• Artists leveraging AI in creation, production, song-writing, audio-engineering, etc.

• AI song generation platforms (Suno, et al)

Maybe AI in the music stack simply sits elsewhere than the DSPs, with value leaking from their walled gardens.

Then-again, they may simply go with a Second Mover strategy: sidestep the legal landmines and let some disruptive player navigate the headaches to establish a market, then either buy or build into it in classic Apple fashion.

This core business question could certainly be motivating (in part) their emphasis on introducing video as well as their rolling up of other audio content, namely podcasts and audiobooks which drive significantly higher margins.

All said, I'm a huge fan of the platform and optimistic for the future, and excited to see how larger brains than mine grapple with this conundrum in real time.

// 🍋

Maybe I’ll get my writing back on track. Maybe not.

We’ll see.

// 🍋

Subscribe now

The longer I work with startups, the more acutely I focus on distribution. It isn’t enough to have a great product if you can’t get it into the hands of your customers. The same goes for ideas. A lone genius in a village has little impact if her ideas can’t spread.

Some of humanity’s most impactful leaps forward have been accelerants for the distribution of information. The printing press and the internet (by extension, social media) were major upheavals for the interchange of ideas. If we think in terms of information as water and distribution as channels, these two inventions were equivalent to the smashing of dams. Each accelerated the flow of information in unimaginable ways.

Extending the metaphor, distribution widens the channels, but in order for rivers of information to flow there needs to be volumes of content pushing the current. So, what taps the source?

Creativity is the wellspring of ideas, and there have been a few brilliant technologies that have unlocked it at scale.

The best examples recently bend what I think of as the ‘arc of accessibility’. It’s the idea that consumer technology broadly arcs toward democratizing access, and some of the most successful startups accomplish this aim. For example, AirBnB’s and Uber’s breakout successes were in part attributable to their ability to capitalize on demand for premium experiences (accessible travel lodging, cheap personal drivers), while tapping latent resources (extra rooms, individuals’ cars). As it pertains to information, there have been a number of technology, gaming, and media startups that have risen to prominence by unlocking latent human creativity to empower individuals while creating content at scale. I believe that the explosive emergence of agentic coding platforms is the next generation of these disruptors.

Unlocking Creativity at Scale

Before speculating on where we are going, it’s important to consider where we’ve come from. The printing press and the internet fueled broad macro trends, unlocking myriad technological advancements. But as we approach the present bleeding edge, let’s consider some micro examples of what I really mean. A couple technologies that sit within my framework for unlocking creativity at scale are digital audio workstations (DAWs) and user-generated content (UGC) gaming platforms (i.e. Roblox).

Digital Audio Workstations (DAWs)

Digital Audio Workstations (DAWs) are programs used to record, edit, and produce music. These pieces of software are incredibly complex and computationally intensive. For context, producing a single track may require a number of individually taxing workstreams to be running in parallel (virtual instruments, data-heavy audio, effects, plug-ins, sub tracks, etc.). Hence, these tools weren’t really commercially accessible until the late nineties and early two thousands as the power of personal computers increased exponentially.

DAWs’ mass market viability rode the wave of Moore’s law, and their proliferation changed the game for musicians. These pieces of software significantly decreased the cost of music production while simultaneously expanding access to studio-grade tooling. It’s hard to quantify directly, given the numerous inputs in producing and selling an album (recording, mastering, mixing, etc.), but I think it’s directionally informative to note that Wikipedia’s list of most expensive albums are largely clustered around the time period right before digital music tooling was taking off (‘90s-’00s). The popularization of programs like Ableton, Logic Pro, and FL Studio empowered independent producers by dramatically reducing the cost to create commercially-viable sounds.

Historically, an emerging band had a minefield of costs to navigate in creating their first radio-ready tracks. These costs could have included limited and expensive studio time, audio engineering, mixing/mastering, and other nickel and diming across the value chain, not to mention the pains of physical distribution. With the advent of DAWs, a single skilled creator could produce a (roughly) commercially viable track in his bedroom at a fraction of the cost, a phenomenon especially potent in more digital-native genres like House music. This widened the aperture for aspiring creatives, and was supercharged by the internet’s distribution channels.

DAWs developed in tandem with the birth of the internet, which upended music discovery, sharing, and listening. Permissionless distribution like Soundcloud and early music blogs transformed virality in a new way, as emerging artists like Odd Future were far less beholden to label strangleholds on physical distribution. This has continued to evolve today as startups push the bounds of distribution as well as real-time collaboration (which has historically been a challenge with DAWs). A great recent example of this is [untitled], which Packy McCormick dissected at length.

UGC Gaming Platforms

If you’re outside gaming, it’s easy to overlook Roblox and Minecraft as simplistic toys for childhood entertainment. It’s underappreciated just how massive these businesses are and their importance in profitably pioneering UGC in gaming. For scope, Minecraft is the best selling game of all time with over 300M copies sold (DAU undisclosed) and Roblox reportedly clears over 70M daily active users. But what’s driven their virality and how did they unlock creativity?

Both platforms were launched in the early ‘00s and center around lego-like functionality where the ‘games’ themselves are sandbox style play environments. Players enter the worlds, and then have free reign to collect resources and build to their hearts’ content. Where most traditional games follow pre-ordained stories with defined checkpoints and goals, sandbox games like Minecraft simply drop players into a blank slate world and provide them with the tools and resources to entertain themselves. The open-endedness of the platforms leaves it to the user to decide how shallow or deeply they would like to engage.

For the novice player, often younger demographics, these games can be played as simply as “gather wood, build house, kill monster” and offer basic play patterns to mete out a meager existence mining or farming. But where they really struck virality is in the complexity that can be built thereafter. Beyond basic resources, the games offer deeper mechanics, enabling players to create mechanical and electrical components like pistons and circuits to build increasingly inventive contraptions:

The early inventiveness of their users, in combination with burgeoning video-sharing platforms like Youtube, helped to drive the viral adoption of Roblox and Minecraft. These blank slate game worlds provided simple, interactive environments in which players could express their own creativity, building everything from villages to assembly lines.

Roblox took this a step further with the introduction of the Roblox Developer’s Environment (now Roblox Studio) in 2006. This developer program allowed players to go beyond ‘in-game’ creation within pre-programmed sandbox levels or simply playing through content created by the Roblox team, and empowered independent developers to actually create their own games for the broader Roblox ecosystem.

But where it gets really interesting is in the monetizability of the games themselves. In 2013, Roblox introduced DevEx, which allowed developers to exchange the in-game currency (Robux) for real-world dollars, thus monetizing their on-platform creations. This supercharged developer adoption by aligning incentives between Roblox and its creator community, allowing independent developers to capture a slice of the value they contributed to the game’s ecosystem. Adept monetization created a massive flywheel effect, which was kickstarted by the already-captive audience on the platform, and has been highly lucrative for both developers and Roblox:

These two platforms have been trojan horses for unlocking gamer creativity at scale. Gamers of the early aughts would have needed coding expertise and heavy-duty computers in order to program games. Minecraft and Roblox empower amateurs to learn and build in a vastly more accessible way. Where a bedroom game developer would previously have had to code an entire game and figure out distribution, today they can learn in a seamless, self-contained platform like Roblox which also contains built-in distribution at massive scale (that 70M DAU we mentioned earlier). This spins a powerful flywheel for emerging and professional developers alike to hone their skills on a captive audience.

Interestingly, the gamified engagement that these platforms created has captured attention beyond simply recreation:

These two games have made learning basic programming, chemistry, and physics more enjoyable experiences in a way that educators have been attempting to crack for years.

Simply, these technologies both bent the arc of accessibility for everyone from hobbyists to professionals. The same leap forward is happening within software development today.

AI and Vibe Coding

For those of you that haven’t followed the AI arms race, things are getting pretty wild. ChatGPT and other LLMs were initially exciting to the average joe (read: me) for their robust search and chat functions. But where things are really exploding is the exponential improvement in agentic coding (coined ✨vibe coding✨). That is, the ability for AI to turn the text of your ideas into iterations of software.

Platforms like Replit, Lovable, and Cursor enable the utterly untechnical to go from vision to prototype with a mere penning of ideas. They have burst open the bastion of software development in a way that is hard to fathom. Historically, devs have been highly sought after and handsomely rewarded precisely for the scarcity of their expertise. The launch of these vibe coding platforms has created a reality where an unskilled troglodyte like myself can code with the efficacy of a first year comp sci student, all from the riffing of a few ideas in a text box. Sure, I can’t magic away the nuance and systems skill of a battle-hardened dev (yet), but I can whip up a shitty MVP of almost any idea in the span of a few hours. This technology has enabled the non-technical to become 1x engineers, and the highly adept to become 100x engineers. And, at risk of sounding over-exuberant here, you have to realize that these tools today are the worst they will ever be. It only gets better. Exponentially.

For example, I’ve had a passing interest in learning rhetoric. I minored in English Literature in undergrad, and one of my biggest lamentations is that I was never able to fit a Philosophy 101 class into my overburdened schedule. Hence, I’ve never had any formal education on the various schools of thought, and my love for Greco-Roman history always made me wonder what it would be like to learn the art of argumentation in a structured way.

Anyway, I’ve had that idea cooking as the ChatGPTs of the world began to improve their agents, and their personalities. But what really urged me to take the leap and build something for myself was Twitter. The longer you swim in that cess-pool of ideas, the more idiotic and vapid you realize the absolute clownery of discourse is there. Flawed reasoning and fallacies abound, which made me wonder: “Could I create an app or agent to teach me rhetoric like Cicero?”

A perfect little use case for testing these tools, and they did not disappoint. In few hours one Sunday, I was able to whip up a janky ChatGPT wrapper that would prompt me with debate topics, and provide a sparring partner and coach in a fairly robust way:

For the flow itself, Cicero’s a fairly simplistic little web app. It creates some randomized debate topics, then it’s effectively just an agent tapping ChatGPT’s API as it ingests your thoughts and itself responds (as prompted) like a rhetoric teacher of old. Like anything LLM related, it took a decent bit of prompt-crafting to refine exactly how Cicero surfaces debate topics (“Make them less dry, make it more topical, funnier, etc.”) as well as how it goes through the motions of debating/coaching you:

But what’s equally magical to me are the little things. The devil’s in the details around my little chat bot. Adding a timer with a corresponding visual bar at the bottom? Half a second. Generate a pixelated pantheon wallpaper? Half a half a second. Completely rework the flow to include back-and-forth with the Agent as well as better coaching characteristics? An abysmal three seconds.

Then you start cross-coordinating your AI’s..

“Hey ChatGPT, provide me with a detailed, structured outline of how to build ABC project. Below is my vision including functionality, aesthetic, and user flow. Please structure your response to be pasted directly into Replit.”

I know it’s a meme– but you really are limited only by your imagination and your agency. Where there’s a will, there used to be a way. Now, where there’s a half-baked stoner idea at 3AM, there’s an MVP. The cost to ship is collapsing, and it’s empowering an entirely new generation of solopreneurs and indie hackers. What the printing press did for the proliferation of reading and writing, vibe coding might just accomplish for software development.

This all converges on a thesis I stole from my chats with Brexton, the founder of Ohara: software will become content. The channels of distribution remain established, and they will soon be flooded with software not unlike the proliferation of video from smart phones or books from the printing press. Personalized apps, software as a meme, and any number of inventive applications may come out of this, with value accruing to a Youtube-like player (or perhaps more closely a Roblox-like behemoth) that can aggregate the creation within the confines of their own platform.

Conclusion

History doesn’t repeat itself, but it does rhyme. Making that which was rare accessible to the masses is both a means of fomenting tectonic cultural change (think: literacy, internet) and accruing unimaginable personal riches (think: YouTube, Roblox). The incentives are ripe for vibe coding to follow a similar path to previous creativity breakthroughs.

The dam has cracked. What was once the domain of highly specialized experts is rapidly becoming a playground for anyone with an idea and a keyboard. As software becomes content, the ability to create will slowly shift from a privilege to an expectation. In the words of William Gibson, “The future is already here– it’s just not evenly distributed.” and the opportunity is immense for those who seize it.

- 🍋

Addendum:

I wrote the first half of this article on DAWs and UGC a few months back, then the back half as I played with Replit. It’s been a couple weeks since I got around to publishing the piece, and it’s already outdated (of course).

A few bad ass examples of precisely what I mean:

1) We may be seeing the very evolution of DAWs themselves right before our eyes. The birth of Model Context Protocols (MCPs – a coordination layer between LLMs and other software) has empowered vibe production akin to vibe coding:

2) About a month ago, Levelsio (a famous indie hacker/solopreneur) went giga-viral with one of the first vibe coded video games: fly.pieter.com (thread)

Twitter lost its mind in the ensuing mayhem as he raked in cash, monetizing on various purchasable plane types as well as ad space, and exceeding $90K in the first month:

The thread goes deep, and was a fascinating saga.

Suffice to say, these tools are going exponential. Software will become content.

Soon, the only real excuse for not creating will be a lack of agency or imagination. Start playing around.

Subscribe now

Last weekend, a generation of creators learned a painful lesson about platform risk (for about 12 hours). The TikTok debates about legality, morality, national interest, and youthful brainwashing will continue to rage long into Trump’s presidency. But, amidst the vitriol and uninformed Boomer takes on the timeline, a more interesting aspect of the drama was surprisingly underdiscussed:

To which, many will respond “Oh, boo hoo, a bunch of Fortnite dancers and makeup tutorial zoomers will have to get a real job now. Woe is me…”, so I think it’s important to take a second to appreciate just how impactful TikTok has been and just how wide its reach is.

I’ve dissected the means by which TikTok has emerged as an Alpha Predator at length here, so I won’t bore you with the gory details. But suffice to say that TikTok’s ability to connect any creator, small business, brand, or startup with their precise audience is unparalleled. It’s incredible. Tiktok knows my taste in music better than I know myself– and it serves up the heat endlessly.

This is evident in the numbers, as the platform has increasingly wolfed down market share:

It’s a better platform for niche creators, but it’s also a beast at scale. That sauce equates to a massive ROI on content and ad spend that startups, creators, and small businesses simply can’t find elsewhere. Particularly after Apple’s introduction of App Tracking Transparency (ATT) in April 2021, which hamstrung Meta’s (Facebook + Instagram) entire advertising apparatus by limiting its access to high fidelity user data. This has made it increasingly challenging to establish distribution (higher CAC, lower ROI on ad spend) for under-resourced players in the consumer space. More than that, on a very human level, TikTok has empowered any creator gritty enough to feed the algorithm with a seamless means to find his or her audience in a way that other platforms lack. This is why it’s been a darling for both brands and individuals alike.

As an aside- I’m not defending the platform as immune to political pressure and it very likely is a piece on the board of geopolitical propaganda. TikTok should be divested, and I generally think that it’s absurd to allow ByteDance (TikTok’s parent) unfettered access to U.S. markets when the U.S. tech powerhouses are unanimously blocked by the Chinese Firewall.

But the point here, really, is that TikTok has created a better mousetrap to connect humans at scale, and the generation of creators, startups, and brands who have devoted their blood, sweat, and tears to establishing the bedrock of their communities on that network experienced a rude awakening last week– the pain of walled gardens and the pitfalls of building a foundation on shifting sands.

Walled Gardens

Two thoughts that have been marinating in my skull:

1. How is it possible that, in the year of our lord 2025, creators still have virtually zero ownership over their audiences?

And, as much as it begrudges me to say this,

2. Crypto actually fixes this.

“Wynn, you mean the shitcoin infrastructure that just empowered our Grifter in Chief to make a mockery of the highest office in the land while simultaneously (and unabashedly) enriching himself and his family?”

Well… unfortunately, yes. But you must believe that, like any technology, the underlying is amoral. It’s the means and ends for which it’s used.

Allow me to explain.

When you think about it, it’s kind of absurd that the largest artist in the world, let’s say Taylor Swift, going to a new social media network (say, launching her Twitter presence) starts at 0 fans just the same as your conspiratorial Uncle who created an X account to ‘fight woke’. But maybe Taylor’s not the best example, because her people will find her. Let’s take it’s murph, one of my favorite emerging DJs. Sure, he’s got it on Spotify and Instagram, but when he launches on a de novo platform, he has virtually zero visibility. He has to shill repeatedly on every other Instagram post “Yo, I’m on X now, follow me on all platforms for show updates.” in order to cross-pollinate his audience and actually accrue value across all networks.

The fragmented nature of one’s audience is a minor nuisance for Taylor Swift’s marketing team, but it could be the difference of making rent for the starving artist who’s making ends meet with a couple of these creator payouts. I saw this first hand when I dated a Twitch streamer who made a healthy living on that platform, but who had to slog away for months to establish herself on Youtube as well. It’s a frustrating reality that brands and creators have to navigate engaging with their arbitrarily fractured audiences.

And in a post-Twitter (now X) world, your very livelihood, the success of your business or the runway of your startup, could be held hostage by the capricious whims of a technocrat or the everpresent shifts of the algorithm itself. We saw the self-interest of these walled gardens on full display when Elon acquired X and went to war shadow blocking and deprioritizing Substack links, which was on its way to becoming the de facto platform for long form content. I analyzed that at the time here, and Packy McCormick had many similar thoughts as well.

Most veteran creators know this. They’re wise to the game and there’s a very real reason that they all have a newsletter and (annoyingly) funnel you to it. Email is open and permissionless. Once a creator has your address, they’re offered an unmediated channel between themselves and their audience as well as the ability to switch mailing services frictionlessly. This kind of control simply doesn’t exist within traditional social media networks where the underlying platforms are highly incentivized to lock in both creators and audiences, at the expense of the freedoms of those very users.

Crypto Fixes This

But how does the technology powering your daily Trump family shitcoin actually stand to provide creator autonomy?

An Extremely Non-Comprehensive Blockchain Overview

At its simplest, blockchains like Bitcoin represent decentralized ledgers, with a network of computers all working in unison to validate the record. This is interesting for basic accounting functions like payments. Then came along next-gen chains like Ethereum and Solana which built smart contract functionality into the fabric of their software. This meant that next-gen blockchains of that time (the early teens) could be more than simple ledgers, but could act as decentralized computers, running more complex and varied programs whose state was verified by the network’s decentralized node base. What this has come to mean in practice is that, through these more performant blockchains, a global network of developers now has access to permissionless, shared infrastructure on top of which anyone can build, iterate, and collaborate. With smart contracts, a world of apps can be built on crypto rails– everything from financial applications like lending platforms to social protocols that track users and record content. The interoperable nature of blockchains means that anyone building in these ecosystems can tap the global pool of users and leverage the work of other projects built on chain. This is to say nothing of the added ability to coordinate capital through token incentives, which is beyond our scope here.

How This Could Be Applied to Social Media

A crypto-native ecosystem could empower creators in three key ways:

1. Audience ownership

2. Data empowerment

3. Financializing followings

Audience ownership and data empowerment is driven by the open nature of the data recorded by decentralized blockchains. For simplicity, interactions on chain are performed through a user’s wallet. As blockchains are distributed ledgers, the various interactions between wallets are public and verifiable. For the crypto-novice, you could think of a wallet address as something like an email address with financial capabilities, and in fact there are services to index and reserve your naming real estate (ex: ENS - Ethereum Naming Service) akin to domains on the traditional web.

Creators and builders in a crypto native ecosystem can capitalize on these open ledgers in a couple of ways. The blockchain records any given transaction between wallets, whether that be between individual users or the wallets of the apps themselves. What this means is that interactions between Wynn.eth and TSwift.eth (in the case of perhaps following Taylor’s account on a Web3 Spotify or Twitter) are open and available. The accessibility of this data allows Taylor Swift and team to see the various addresses that have interacted with her content. This is valuable in seeing frequency and depth of my interactions with her content, as well as similarities across other artists with whom I’ve interacted. But more than this, the team as well as other builders now have a higher fidelity view of who and how these wallets have interacted across her fan universe.

So, a competing social media app (let’s say a Web3 Twitter) could scrape this data for each new user that comes to its platform. When TSwift creates her Web3 Twitter account, she could automatically have the new platform push something to the fans of hers that had previously interacted with her wallet on Spotify (artist to fan). Or alternatively, I could have the Web3 Twitter scrape my wallet history for Spotify-relevant transactions with corresponding artist wallets that have since created accounts on their platform, and apply some rules-based approach to follow the artists with whom I’ve most interacted (fan to artist).

These data advantages can be sliced and diced hundreds of ways, many of which were experiments tested in the last cycle’s cambrian NFT explosion. We saw artists like Metallica drop NFTs for their super fans, thus providing the band with visibility into their fan base as well as ‘willingness to spend’ of super fans. We saw a swarm of startups attempt to disrupt the vampiric toll extracting TicketMasters of the world by bridging direct relationships between artists selling tickets and NFTs dropped directly into fans’ wallets. There was also an entire throughline of thought around curation and ‘proof of fandom’ where early superfans and music tastemakers now have verifiable demarcations of their curation in spotting up-and-coming artists, and you could envision artists later in their career rewarding their earliest fans (which they could easily identify down the line via historical interactions on chain).

Crypto continues to be a promising creator technology in so many ways, and this type of data and unmediated access has had technologists, creators, and artists foaming at the mouth for years now. Unfortunately, Web3 has largely failed to deliver on these ideas for a number of reasons– immaturity of the technology, lack of infrastructure, over-financialization, etc. That said, tech progress comes in fits and starts, and with each hype cycle the space has continued to serve up new attempts at revolutionizing the creator economy.

Who’s Trying Today

Farcaster is a startup founded in 2020 by Dan Romero and Varun Srinivasan, two early employees at Coinbase and successful founders in their own right. The team has been building relentlessly for four years now, and they’ve seen some success, accruing $150M in funding from luminaries such as USV and a16z Crypto as well as peaking at around 80K DAU on their client Warpcaster:

But what is Farcaster?

Farcaster is a sufficiently decentralized social media stack being developed on blockchain rails.

That’s a lot of buzz words. Let’s unpack.

The organization of the network functions through two layers: the protocol (Farcaster) and the client (Warpcaster).

The protocol layer (Farcaster) is comprised of a collection of decentralized nodes (Hubs) that coordinate to record the data of the network. The nodes themselves are permissionless and independent such that communications cannot be arbitrarily censored and the data is openly accessible.

An aspect of the ‘sufficient’ caveat comes from the way that Farcaster is constructed, with a blend of on chain and off chain data. Blockchains, even the most performant, haven’t yet reached the point where it would be economically viable to store exclusively on chain the petabytes of data a social media network generates at scale, so the founding team has made structural compromises around what type of things should be immutably stored and what will be owned locally at the client/app level:

“The on-chain aspect of Farcaster is limited to identity; every user mints a unique Farcaster ID NFT (FID) to participate. This is their permanent ID in the Farcaster protocol. Users sign messages with their keys, mapping all their content to their FID. As FID lives on Ethereum it is censorship resistant.

The actual content (ie. the social graph) is stored in the off-chain Farcaster network run by Hubs. Hubs are like Ethereum nodes. Anyone can run a Hub. Hubs gossip messages to each other and frequently sync to have a consistent view of the network.” - Can Gurel, Farcaster in a Nutshell

Said another way:

“A social network achieves sufficient decentralization if two users can find each other and communicate, even if the rest of the network wants to prevent it… [which] can only be true if developers can build many clients on the network.” - Varun Srinivasan, Sufficient Decentralization for Social Networks

The team has built the ecosystem in such a way that it is credibly decentralized, with third party builders and users resting assured that their social graphs will not be arbitrarily compromised by the network.

Simply, “... the architecture decouples the data itself from the distribution. Hubs store and maintain the data. Apps decide on how they distribute it to the users.” – Can Gurel, When Farcaster Clicks

In the case of Farcaster the startup, the team has developed its own client (Warpcaster) which essentially acts as the front end through which users tap into the Farcaster (protocol) network. Akin to the way web browsers (Chrome, Firefox, etc.) interface with the internet, the content all exists at the permissionless protocol layer and then independent users can build any number of clients to interact with the Farcaster data in their preferred way. Today, Warpcaster looks much like Twitter meets Reddit, with a few crypto-native functions (like Frames, below).

In practice, this shifts the paradigm of how social media platforms compete, pushing innovation to the edges of the network. Dan Romero’s vision has been an open social ecosystem that is BYOC (bring your own client) and BYOA (bring your own algorithm), and it’s frequently a clapback of his to criticisms of Warpcaster’s functionality (“Don’t like our client? Build your own.”). If an individual or community is unhappy with Warpcaster’s content moderation policy or what its algorithms prioritize, the market has full autonomy to pounce on these weaknesses and develop a competing client. This kind of ‘forking’ can happen extremely rapidly, given that any new code base can pull all of the same network data as the incumbent. It’s a market with (near) perfect competition in which users and liquidity are fluid and clients must compete on vectors like UX and functionality in order to aggregate demand.

For example, Alana Levin enumerated a number of dimensions around which various clients could specialize to differentiate:

“If we anticipate a world where there are many engaging Farcaster clients, the question becomes, how do we get there? One route is for developers to directly build new clients: find a media type that isn’t a first-party citizen and build a client to give it a first-class experience. One could argue that this is how we saw popular apps in web2 find fit, from Instagram (images) to Twitter (micro-blogging) to YouTube (video) to Vine (very short-form video). Maybe there will be a similar pattern with new Farcaster clients as well — but I’d look for the emergent data types (like assets or Frames) as opposed to media types that developers have had years to iterate on. Twitter started out as micro-blogging but is now multimedia. Same with YouTube, Spotify, and many others. Thus, the unlock is probably in the new stuff: a first-party client for a data type folks haven’t yet mastered.” - Alana Levin, New Opportunities in the Farcaster Stack

Another aspect of the green space created by building a social network on crypto rails is the ability to integrate in-app payments seamlessly. Each user within the Farcaster network, regardless of client, is interacting with a wallet living on fully functional financial rails. This allows the interchange of creator payments, tokens, NFTs, etc. to be built directly into the feed, app, or client in a more native way than traditional social media platforms like Facebook or Twitter. And in fact one of the most promising permutations within Warpcaster has been the introduction of Frames, which allow users to build miniature apps directly within their casts (think ‘tweets’) on Warpcaster. Frames have been created for everything from in-cast payments to collaborative gaming:

I fully anticipate Frames to drive the next leg up in the Farcaster ecosystem. The cost to ship light-weight, tiny apps is collapsing in a text-to-app powered AI world. Warpcaster’s design is uniquely situated to accommodate this in a way that many of the legacy social media titans are not.

Conclusion

The beauty of the internet is that it’s a living, ever-evolving network. Social media platforms of today are far cries from the MySpaces and AIMs that many of us grew up on. As these networks have become increasingly intertwined with the fabric of our society, the questions around ownership, censorship, and distribution have only gained importance. Elon has clearly identified this with his crusade to reorient Twitter as the global townsquare. In many ways, these last couple U.S. election cycles (deplatforming sitting presidents, prioritizing single party agendas, etc.) were a testament of their power and a prime exemplifier of the need for decentralized ownership.

The generation-defining attract-extract model of centralized social media continues to unfold in unanticipated and painful ways for the users of these walled gardens as well as the various startups that have attempted to build on top of them. It is yet to be seen whether crypto rails are the panacea or not. Regardless, in the face of absolute control and shifting sands beneath, the arc of empowerment should ultimately bend toward the user.

- 🍋

Subscribe now

The Acquired podcast changed my perspective on Meta’s investment in Reality Labs (AR/VR):

“I'm pretty sure glasses and holographic presence and AR is going to be a completely ubiquitous product, right, it's just like everyone who had a phone before replaced it with a smartphone and then a lot of more people got smartphones. If all we get is all the people in the world who already have glasses upgrading to glasses that have AI in them, then this is already going to be one of the most successful products in the history of the world, and I think it's going to go a lot further than that. So I know there's that.

There is the thing about controlling our own destiny. It's strategically valuable. You know we did this calculation or estimate at some point where it's like how much money do we lose from our core family of apps to the various taxes that the platforms [Apple’s App Store, Google, etc.] have to like, when they tell us we can't run the ad business the way that we think we should be able to.

When they tell us we can't ship certain products, that way people use the things less or like them less, and it's hard to exactly estimate it. But I think we might be twice as profitable if we own the platform or something. I think from that perspective that's worth a lot, just from a pure dollars perspective, which is not primarily how I come at this stuff…” - Mark Zuckerberg on Acquired

AR/VR was all the hype around the time that Zuckerberg rebranded Facebook to Meta, and Wall Street torched him in the ensuing years for the company’s massive investments into augmented reality technology as the sector fell out of favor. But when you frame it within the context of owning the next platform, their hefty R&D bill starts to make a bit more sense. You can basically run a loose NPV for what that’s worth against the cash sink to develop the technology today. This theoretical number could be so huge that it’s rational to continue your investment as an option on a gargantuan future market. Whether or not AR/VR becomes the dominant platform is anyone’s guess, but if you’re the founder and controlling shareholder of Meta, and your timelines are measured in decades, there is a serious logic in betting on (and willing into existence) augmented reality if the probability of its ubiquity is high enough.

This is the power of platforms.

Owning your own destiny has proven time and again to be fundamental in cementing network effects with hardware and user lock-in within your ecosystem. It’s tech strategy 101.

Historically (shooting from the hip here):

• Microsoft: Had significant market power as the dominant operating system, and integrated its products to both build and capitalize upon network effects. When the internet began to emerge and Netscape was gaining traction as the go-to web browser, Microsoft simply bundled Internet Explorer with its operating system to quash the competition. This type of bundling strategy was so valuable/restrictive that it ultimately led to the company’s antitrust woes in the 90’s.

• Apple: Called (and revolutionized) mobile perfectly, and has been able to leverage that ownership ever since. Seamless integration between your iPhone, Airpods, and Macbook creates a hardware ecosystem, and dominating the app store across all these platforms to extract onerous fees indefinitely (See: Epic Games vs. Apple) monetizes the software.

• Google: A search engine behemoth saw the mobile revolution, and ensured that it had a horse in the race, reducing its reliance on and gatekeeping by Apple, by acquiring Android. Android has since become the largest smartphone platform today, natively preferencing Google’s product suite (Gmail, Maps, etc.) and helping to proselytize Chrome much like Microsoft’s play before it.

So, platforms are imperative. And beyond consumer hardware (AR/VR), who’s building the next one?

I think the framework above is an interesting lens to contemplate the humanoid robotics landscape. It adds some meat to the various investments and partnerships we’ve seen:

• Figure: partnership with OpenAI to develop humanoid AI models + $675M round from Microsoft, OpenAI Startup Fund, NVIDIA, Intel, Bezos Expeditions.

• Rainbow Robotics: Samsung’s $181M investment for 35% ownership.

• Apptronik: partnership with Google DeepMind.

• Agility Robotics: investment from Amazon’s Innovation Fund.

• Tesla: speculation around collaboration with X/xAI (supercluster capabilities).

If you consider humanoids the next big wave, however far out, then it is imperative to all the major players to have a ‘piece on the board’ if even to just maintain option value.

This is what came to mind when Apptronik announced its partnership with Google’s DeepMind in late December. I’m sure that Google did an incredible amount of diligence on the various players in the space and likely had carte blanche in their choice of partner (being DeepMind and all), so I’m not trivializing Apptronik’s state of the art technology. But I use this to say that many of the forefront competitors had already been tapped by the other big dogs in AI (OpenAI, etc.), and Google didn’t have a horse in the race. The major tech companies all want a piece on the board, if only to keep a pulse on the industry and maintain an option.

Additionally, the import of these options compounds if you believe that value will ultimately accrue to the intelligence layer (higher margin software). It’s the same playbook as the infra providers investing in LLMs, a la Microsoft <> OpenAI. Sure, they’ve invested significant sums into OpenAI– revenue share aside, and much of the investment was actually compute services instead of cash– but that’s a small price to pay to get ingrained into a potentially massive future customer. All of the protocol level LLM companies are going to require massive amounts of compute, which the hyperscalers are all competing to provide.

I fully expect a parallel with humanoids. Yes, OpenAI is investing and learning and having a dog in the fight. But they’re also securing themselves in pole position to be the vendor of choice for a potentially massive customer (Figure). Embodied intelligence will need just that: intelligence. The value likely accrues to the intelligence layer, and all the providers of that service are happy to ingrain themselves with the corresponding hardware builders.

How this plays out in practice is an open question, to be fair. Do the humanoids themselves eventually become commoditized, with the ‘sauce’ being the interchangeable AI that powers them? I’m sure it’s likely to be a spectrum of strategies. All of the robotics manufacturers will have some level of AI expertise to be sure, but the massive data processing required to train and perfect models across fleets at scale may lend itself to specialization in the value chain. For example, DeepMind might be the big data + model provider for Apptronik’s edge compute + hardware expertise. Then there will likely be other players that run a vertically integrated strategy, which seems to be Tesla’s approach leveraging its in-house expertise from autonomous driving.

The race has just begun, and I’m eager to see how it plays out. Admitting my ignorance to the particulars of the space, I’m sure that much of what I’ve said here could turn out to be completely off base. Nonetheless, I think that the platform framework helps to rationalize the spend from Big Tech as they look to capitalize on one of the potentially huge next waves.

- 🍋

Subscribe now

Robots are getting wild. We’ve got Boston Dynamics bots doing backflips, Tesla Optimuses DJ’ing launch parties, and don’t even get me started on the Thermonator..

Yet, despite all these advancements, robot integration into our daily lives has been slow to proliferate. Driverless cars are probably the most frequent manifestation of autonomy I see here in Austin on a daily basis, but little beyond that. Sure, we may eventually see an army of humanoids scurrying around cities solving our every whim, but that future appears to be a ways off. In the meantime, there are bathrooms to clean, floors to vacuum, and lawns to mow.

The stories in the press converge on the flashiest, sci-fi use cases (see: above), but it’s an open question as to what everyday autonomy will actually look like. Is it drone deliveries, autonomous skiffs, driverless cars or humanoid man-servants? Likely all of the above, eventually.

One framing of this debate is whether we see generalized humanoids or specialized alternatives. Billions have been poured into generalized humanoids, like Tesla’s Optimus, Apptronik’s Apollo, and Figure’s Figure 02, that promise to revolutionize everything– heavy industry, space exploration, mining, energy, you name it. This has produced quite a bit of hype.

But in the meantime, there’s an entire ecosystem of brass tacks roboticists being largely overlooked. Some of the coolest bots I’ve seen recently are being built by founders automating dirty, taxing jobs to drive a cleaner world, and I think there may be a market opportunity for a specialized platform player to emerge while the humanoid alternatives are still learning to handle a broom.

Let’s get our hands dirty.

Bathrooms, Carpets, and Lawns- Oh my!

Somatic

Autonomously cleaning a bathroom is a deceptive task. At first blush, if you’re like me, you’d imagine this routine chore to be relatively easy: scrub floor, clean toilet, rinse sink. It’s something that we humans (begrudgingly) handle with ease.

But then you consider the complexity required to actually automate such a task. You need your little cleanbot to open/close creaky stalls, unlock confusing door latches, navigate varying urinal and toilet configurations, spray different sink shapes/sizes/heights, mop pitched floors, and handle a hundred other minute details. All of which must be done at a speed, quality, and cost that’s competitive with today’s janitorial staff. It’s no small feat, but this is exactly what Somatic has done.

The YC-backed company is the brainchild of Michael Levy and Eugene Zasoba. The team have been partners in crime for over a decade, having previously scaled and sold edtech SaaS platform Chalkable in 2016. Michael has said that his scar tissue from the edtech space spurred him to look for precisely the opposite type of business next: no regulation, no competitors, and anything where tech giants like Apple, Google, and Facebook weren’t building. So, he started playing with robots.

His first foray into the space began with Trike, an autonomous wheeled delivery drone. It was a three year project that got the team building bots, and Michael points to this as their education in hardware. That work helped the team navigate the mind maze of where they could really find a market to win while providing them with a thorough understanding of what it takes to build bots. Ultimately, although package delivery is huge, it’s a crowded market and a highly variable problem set with constantly shifting routes, legions of competitors, and all manner of unexpected obstacles.

So, they turned to toilets.

Set piece tasks on consistent schedules within stagnant floorplans. Simple! Minus the myriad complexities listed above…

Even if you build the bot that can do it, working out the economics is a challenge in and of itself. The machine must be as quick and thorough as a human janitor for a fraction of the cost, all while being a seamless out-of-the-box solution for maintenance coordinators to implement. This gets to the heart of Somatic’s value proposition.

After five years of iteration, the startup has proven two major milestones:

1. Not only can its bots meet the cleanliness standards, but

2. They can be deployed in a seamless way

Their current machine operates using a robot arm and a number of attachments for its various functions: soap/water spray, dryer detachment, wet vac for floors, and a larger vacuum in the base for broader surface areas. In their most recent update, you can see the Somatic bot go from unplugging itself in the janitor’s closet through cleaning multiple bathrooms completely unaided. It’s taken time to nail commercial viability, but the team has continuously iterated with its customers and have now deployed almost a dozen robots, with ambitions to scale to hundreds in the next 24 months.

But high-touch, white glove pilot customers do not a scalable startup make. The second key to their solution is the actual deployment. The team has engineered their next generation bot to be delivered on-site to a customer, assembled by the local staff, and then programmed and deployed completely remotely. This is fundamental to scaling the Somatic fleet from a handful of machines to hundreds of customer sites in a labor-efficient way.

That said, quality of service and speed of deployment ultimately intersect at the question of cost. Somatic’s pricing is as follows:

A high level cost analysis can be run using Janitors & Cleaners stats from the Bureau of Labor Statistics. With the average annual wage of janitors around $36,250 (assuming average 40hr/week), Somatic’s 8hr weekday option is about 50% cheaper on an annualized basis. Of course, this comparison isn’t truly apples to apples— there are still a number of tasks that will require a human in the loop like refilling paper towels, flooded facilities, etc.

Nonetheless, at this price point, the Somatic bot is an attractive solution for janitorial service companies that face constant challenges with turnover, labor shortages, and quality consistency. I believe that the savings are significant enough to drive early ‘tech-curious’ facilities managers to pilot the Somatic bot, opening an opportunity for the startup to ‘wow’ them and expand their early champion base. Plus, having a Somatic machine on-site can also free up janitorial staff for other tasks like vacuuming...

Tailos

Austin-based Tailos is a home-town hero tackling industrial grade, commercial cleaning with a fleet of autonomous vacuum bots. To call it a Roomba on steroids trivializes what the team has built. Rosie, their machine, is effectively a miniaturized Waymo for all your vacuuming needs.

Full disclosure, the startup is a Capital Factory portfolio company, and we’ve deployed their bots at HQ.

Tailos was originally founded as Maidbot by Micah Green in 2015 during his undergrad at Cornell. A classic dorm room startup, the company was spawned from Micah’s experience working as a room attendant at the Statler Hotel on campus. This inspired him to tackle the labor-intensive problems that plague hospitality: bedding, vacuuming, and cleaning. While he initially envisioned a bot to change bedding, his customer discovery bore out that it was too early and capital intensive to develop a robotic solution to tackle that problem just yet (back in the mid teens). So, he doubled down on an equally pressing pain for hotels: vacuuming.

Micah teamed up with a Georgia Tech Robotics professor to scope the initial prototype. MVP in-hand, the team was able to secure their first angel checks and Micah dropped out of Cornell to go all-in. In the decade since, Rosie has come a long way. Although the initial Rosies could achieve the requisite cleanliness milestones, it took a number of iterations for the team to really nail the ease of use required to drive broader adoption. Micah’s quick to highlight that the average age of hospitality cleaning staff sits around 60 years old, so Tailos had to relentlessly simplify UX and refine its GTM. Before Rosies are adopted to work in concert with maids flipping rooms, the initial beachhead at any given property is often around public spaces like hallways, meeting rooms, and ballrooms, where property managers can gain comfort and confidence in the product while educating their staff.

Today, the team has nailed the customer pain point within hospitality, and has since expanded to other use cases like senior living, commercial real estate, and even cruise ships. The company’s fleet tops 1,000 robots deployed today across 12 countries, and their growth has been fueled by tens of millions in venture funding across multiple rounds. Rosie’s successes have even caught the eye of strategic players like Reckitt, owner of the brand Lysol, who led Tailos’s Series B. But why the interest, and how does a Rosie differ from a simple Roomba?

As you can imagine, the commercial demands for this use case are far more stringent than a simple consumer machine. Rosie isn’t much bigger than a standard Roomba, standing 12x16x6 inches and weighing approximately 18lb, but the platform packs far more of a punch. On the specs, the Rosie is about 10x faster than consumer robots, 5x more powerful, and has a much higher bar for durability (with most users running it 4-8 hours a day). These features are key for hospitality customers whose brands are built on tidiness. Additionally, the speed is fundamental to the ROI equation, where hotel property managers are often deploying Rosies to complement their maid staff in order to turn rooms more quickly (vacuuming while a human changes bedding, etc.) and drive higher throughput on their properties.

But where the business gets really interesting is the tech on top of vacuuming. Tailos machines are also outfitted with a host of sensors and supported by a software stack that further drives utility. On the sensor side, Rosies map where they’ve cleaned, edge cleaning percentages, and other data points like wifi strength, air quality, and mold levels across a property. This gives property managers direct insights into the cadence and cleanliness of their hotels at any given moment. The team has also built out swarm capabilities to coordinate multiple Rosies in real time for cleaning larger surface areas like ballrooms and corridors. This is all underpinned by an extremely simple UX that empowers customers to “press play, walk away”, dropping Rosie into any environment and letting it run autonomously straight out of the box.

But Is It Fundable?

So, the tech is viable and the solution is cool, but the question remains: is it venture-backable?

Beyond business model and capital intensity, a couple broad-based considerations come to mind: defensibility and market sizing.

For the former, there’s a small camp that might point to the hype around humanoids and consider their rapid iteration on a longer timeline. Can these specific use case solutions scale to a meaningful outcome before general purpose counterparts come down the cost curve? This risk factor is fairly easily addressed. Ultimately, I think our future cities will look like Disney’s Robots or Star Wars. Sure, we’ll have humanoid battle droids, but I think we’ll also have fleets of Gonk droids and repair bots scurrying around as well.

Today, general purpose humanoids are advancing exponentially, but they are still more novelty than tradesman. They will first be deployed at the high end for expensive use cases leveraging their unique strengths (3PL/warehousing solutions, manufacturing applications) before cost and scale make them a viable solution for menial tasks like everyday property management. Specialized solutions have a window in the meantime to gain scale in their own right while building out a broad-based platform to land-and-expand within their verticals. This is to say nothing of form factor, where the ideal machine to accomplish something like vacuuming probably isn’t a bipedal pushing a suction stick anyway.

For TAMs, startups like Somatic and Tailos are each going after sizable markets in their own right. But more than singular use cases, I think the real opportunity lies in building a true platform player. Following the Anduril playbook, the biggest winner in this space will build out a suite of machines, all empowered by a unified tech stack on the back end.

Once someone like Tailos owns a property’s vacuuming needs with a sticky product, it’s easy to sell through existing channels with an ever expanding product suite. This becomes a virtuous cycle as your portfolio of bots then coordinates as a network on your software stack.

I think that many investors and founders have come to a similar conclusion. Whoever can scale quickly enough within their chosen beachhead can compound their advantage. First movers will own sticky relationships that will be hard to displace. Once a property manager has an installed base of cleaning bots, there will be friction in switching costs. The leading service provider can then leverage its reliable cash flows to raise increasingly larger rounds to build a product suite in-house or through a roll up strategy, scooping up the subscale players in adjacencies to establish a dominant portfolio. All the while, the frontrunner will be accruing an increasingly differentiated data set on which to build their tech stack, wringing additional efficiencies across their install base as they gain scale.

Conclusion

Despite their lack of flash, these startups beyond the limelite might just be the winners in this robotics hype cycle. Humanoids will be huge… eventually. But the delta between now and viability, let-alone adoption, remains to be seen. All the while, specialist players are deploying in real-time, tackling real world use cases today.

Mountains of venture dollars will be torched in the pursuit of the T800, but it’s an open question which startup will be the Cyberdine and the surrounding hype has driven many of their valuations to far outpace the reality of the technology. Said another way, many techno optimists want shiny humanoids, but the winners for this vintage may just turn out be the Gronk droids along the way.

- 🍋

Subscribe now

SpaceX reignited the space race. With Elon’s success, a second order effect has been the proliferation of satellites in orbit, which has ushered in a golden age of geospatial imaging. The various constellations around Earth today provide rich, affordable, and accessible surface data of virtually the entire world. Not only that, but the exponential decline in launch prices has also inspired scientists to dream bigger and look deeper in the other direction as well, launching mind-blowing instruments like the James Webb Space Telescope in 2021.

So, we’re getting deeper views of space, clearer views of Earth, and yet our view of the ocean remains murky as Galveston Bay (Texas-specific joke, sorry). That’s no small segment of the map…

The ocean is big. Very big. 71% of the Earth’s surface. Yet 80% of the ocean’s floor still remains unmapped. Why?

Because gathering oceanic data is extremely expensive. As with all things, the problem historically was supply and demand. You previously would have had to run manned ships bristling with expensive sensors across large swaths of open sea through treacherous and variable weather conditions in order to build a comprehensive data set including unattractive segments of sea with little customer demand. Highly manual, sometimes dangerous, entirely unprofitable. Not great.

Fortunately (unfortunately), a confluence of factors have unlocked supply while global forces have accelerated demand for all kinds of marine of data:

• Technology - we’ve come down the cost curves in robotics, edge computing, satellite connectivity/availability, solar panel effectiveness and cost, as well as advancements in battery technology

• Global Warming - the effects of global warming have magnified our need to understand the changing oceanic landscape and intensifying weather patterns, and accelerated deep sea mapping projects like Seabed 2030

• Geopolitical Uncertainty - South Seas geopolitical tensions have driven dollars into naval research, data, and readiness for the defense sector

So, we’ve got renewed interest and increasing access, but who’s making it happen?

As We See Above: Saildrone

Saildrone is probably the frontrunner in deploying uncrewed surface vehicles (USVs) for oceanic data collection. And let me tell ya, its Co-founder and CEO is exactly the kind of guy you’d want tackling this challenge. Richard Jenkins has been a lifelong sailor and engineer. He’s been working on ships since his teens, and has been obsessed with the technology of the craft ever since.

But more than that, he’s an absolute legend of a salty sea dog. Prior to founding Saildrone, Jenkins was obsessed with the art of speed sailing. More specifically, he was possessed by the challenge of breaking the land-speed record for a sail-powered vehicle. He devoted a decade of his life to the task, and achieved the record as the ‘fastest sailor on Earth’ in a Las Vegas lakebed in 2009, piloting his faithful Greenbird to 126 mph.

He’s not quite Shackleton, but I like to think they’d get along. Plus, Saildrone 1020 became the first unmanned vehicle to circumnavigate Antarctica. So they’d have plenty to discuss.

Today, Saildrone maintains a fleet approaching 200 vessels and its skiffs have autonomously sailed over 1.6 million nautical miles. The company was founded in 2012 by Richard Jenkins and Dylan Owens (who left the startup in 2014, per his LinkedIn), whose skillsets combined physical shipbuilding know-how with nautical autonomy expertise. The startup’s maiden USV voyage launched in 2013, sailing the first Saildrone prototype, The Honeybadger, from San Francisco to Hawaii over 34 days. Since then, the startup has deployed its fleet globally from the South Seas to Antarctica, and has developed multiple classes of vessel for varying use cases.

The Drones

The Saildrone fleet today is comprised of three models:

• Explorers (15ft): scientific data collection & single-beam ocean mapping

• Voyagers (33ft): maritime domain awareness & ocean mapping

• Surveyors (65ft): deep ocean multibeam mapping

The varying vessels can be outfitted with any number of capabilities:

“Each saildrone carries a payload of science-grade sensors to measure important atmospheric and oceanographic environmental variables in real time, map the seafloor in high resolution, and provide persistent maritime domain awareness. They have been deployed on data collection missions in the Arctic, Atlantic, North Sea, Tropical Pacific, South Pacific, Southern Ocean, Gulf of Mexico, and along the North American West Coast to track fish and mammals, measure how much carbon is being absorbed by the ocean, study why some hurricanes intensify so rapidly, and even stand in for disabled weather buoys.” - Saildrone (source)

The USVs are powered by a combination of solar and wind energy. Interestingly, it was Jenkins’s obsession with land sailing that birthed the innovative sail design that is integral to Saildrone’s edge. The carbon-fiber sail itself is effectively a vertically turned wing, which harvests the surrounding wind to generate ‘lift’ angled as propulsion to power the skiff forward. The odd-looking side structure jutting off the sail is a trim tab, which helps to angle the sail. Furthering the airplane comparison, if the sail itself were the ‘wing’ generating lift (propulsion) the trim tab would be the tail fins of the plane, adjusting via autopilot to steer. This works in concert with a rudder on the ship itself. This system allows the Saildrone units to travel in an incredibly energy-efficient manner, harvesting an infinitely renewable energy source for their movement. It’s also worth noting that this design is mechanically minimalist, which adds to the durability and longevity of the drones and helps to keep down costs.

For the on-board electronics, solar power is captured via panels on the deck and sail of the ships. Leveraging these two renewable power sources enables Saildrone fleets to effectively run in perpetuity. This limited downtime is an obvious boon for driving down the cost of civilian use cases (oceanic mapping, marine life tracking, etc.), but also makes the Saildrone surveillance capabilities particularly appealing for governmental partners as well.

Maritime Domain Awareness (MDA) & Intelligence, Surveillance, and Reconnaissance (ISR)

Saildrone began initial testing with the Navy in December 2021, where it deployed two fleets of Explorers to help monitor surface traffic in the Persian Gulf and Red Sea. From the success of those initial tests, the startup has continued to integrate its unmanned capabilities with armed forces partners. While precise fleet deployment statistics are scant, a few notable data points from Richard Jenkins’s testimony before the House Armed Services Committee a couple months ago (emphasis mine):

“Operational success in the Middle East led to expansion into the SOUTHCOM and NORTHCOM AORs [Areas of Responsibility] last year. There, a fleet of Voyagers has provided maritime ISR for counter trafficking operations continuously since September 2023. Quarterly reviews by the Office of the Undersecretary of Defense for Intelligence and Security (OUDSI) throughout the mission reported that saildrones were responsible for 96% of all surface contact detections in and around the Windward Pass and the Florida Strait with a 95% cost reduction relative to the daily operational cost of a Navy destroyer. Put another way, for the entire daily rate of one destroyer, Saildrone could provide complete maritime ISR coverage along every major Caribbean trafficking route to the United States. As in CENTCOM, our platforms have provided a decisive operational advantage in an area where demand for manned resources is high and availability is scarce. I am pleased to report that we will be doubling that fleet and expanding into INDOPACOM with additional drones this year, enabled through the DIU’s fielding budget line.”

Funding & Future

Venture investors always discuss the proverbial ‘10x solution’ as the kind of breakthrough where startups win. Saildrone has demonstrated precisely this kind of step change improvement over previous labor and cost intensive means of gathering oceanic data. Hence, with strong indications of PMF, the company continues racing to scale its fleet and widen its lead.

The progress thus far has been fueled by and attracted investment from a number of venture luminaries including Mary Meeker’s BOND Capital (a spinout of Kleiner Perkins’s growth fund), Lux Capital, and Tribe Capital, among others. Saildrone secured a $100M Series C in 2021 to continue fueling its goal of quantifying the entire planet.

So Shall We See Below: Seatrec

Unfortunately, not all data points are so easily gathered from the surface. Although USVs are being equipped with increasingly sophisticated sensors, creating a detailed picture of the ocean requires a mosaic of data types, some of which are best gathered at depth.

That’s where Seatrec comes in:

Seatrec has developed a novel technology that leverages gradient changes in oceanic temperatures to generate clean, renewable power for undersea robots. The startup’s modular platform can dramatically enhance (10x+) battery life while simultaneously unlocking new sensor capabilities that had previously been uneconomic to deploy due to their power demands.

One of the primary bottlenecks of deep ocean observation is energy consumption. Where Saildrones can leverage solar and wind to fuel their voyages, bots deployed at depth don’t have that luxury. This makes it incredibly costly to gather deep-sea data– researchers were previously stuck between a rock and a hard place of either paying mountains of cash to manually retrieve and recharge their expensive equipment when it died or losing heaps of pollutive hardware to Davy Jones’s Locker. Either option made this data untenable to gather for generalized use cases. Plus, one can imagine clandestine interests where a surface trawling vessel may not be the ideal medium for intelligence gathering.

The Technology

Seatrec’s technology was developed by Dr. Yi Chao and his team at NASA’s Jet Propulsion Lab over a decade ago (surprise, surprise - NASA is one of the largest gatherers of oceanic observation data). After initial validation, Dr. Chao saw the potential for Seatrec’s solution to dramatically enhance deep sea mapping capabilities, and began leveraging federal grants to hack away at commercializing this innovation. What started as an attachment to extend the life of third party machines has since evolved into the startup’s infiniTE Float, a modular platform that customers can outfit with sensors of their choosing.

But how does it work?

Seatrec is a portmanteau of Sea + Thermal + Recharging. Its underlying technology leverages thermal changes in the ocean to harness energy. Within the float is a small loop containing Phase Change Materials (PCMs), chemicals which easily phase shift from solid-to-liquid/liquid-to-gas, attached to a motor. At varying temperatures, the PCMs transform, changing volume as they expand/contract in various states. These volumetric changes then drive pressurized fluids through a motor which generates electricity. In Seatrec’s case, its internal PCMs freeze/expand as the float descends to cooler, deeper depths and melt/contract as the float surfaces to relay data. This generates enough power to significantly extend float battery life nearly indefinitely, and even unlocks new use cases requiring more power-hungry sensors.

Use Cases

Seatrec’s technology shares a number of similar use cases to Saildrone’s, with a few additional bells and whistles. Seatrec floats can be utilized for seafloor mapping and general data gathering (temperature, acidity, etc.), but they also have unique applications for weather tracking, soundscape monitoring, and potentially UAV recharging.

Both Saildrone and Seatrec technologies are being used to gather meteorological data. As storms intensify and long-tail events occur with more frequency, the ability to disperse tracking technologies across the ocean continues to gain relevance. For example, deploying a fleet of floats across the Gulf of Mexico can empower meteorologists with visibility into surface temperature data that helps predict how much a hurricane might intensify as it approaches land. To that end, Saildrones have been deployed in a number of hurricanes, even sailing directly into the eye of the storms.

One key advantage to Seatrec’s technology is accessing the Sound Fixing and Ranging Channel for acoustic monitoring as well as undersea communication. The SOFAR is a channel around 600 to 1,200 meters beneath the ocean’s surface where sound waves can travel over extremely long distances. This effect occurs due to a unique combination of temperature, pressure, and salinity at that depth.

The SOFAR channel is leveraged by biologists for tracking undersea wildlife like whales, but it also has a number of important use cases for submarines as well. For naval applications, it is used for both listening (intelligence gathering) as well as guidance/communications. Acoustic beacons can be placed at depth as an additional means for submarines to triangulate their position. This channel is also tapped for reconnaissance, listening to the movement of vessels etc. While USVs like Saildrone can certainly eavesdrop into the SOFAR channel via towed hydrophones, constant monitoring from extended life floats like Seatrec’s are a stealthier way to achieve this in more sensitive seas like contested waters.

Finally, Seatrec’s technology is also being explored for some more sci-fi type use cases like underwater power stations or arctic energy bases. In a next generation conflict, autonomous undersea vehicles will be utilized at scale, but powering and recharging these machines remains a challenge. To address this need, Northrop Gruman partnered with Seatrec to test use cases of attaching its power generation technology to Unmanned Underwater Vehicle Stations. It is also speculated that this type of technology may be utilized to power next generation UUV’s like the Navy’s Manta Ray:

Seatrec has also collaborated with the Navy to blueprint power stations in the Arctic, which are largely powered by diesel during times of the year with limited sunlight.

Although its energy output continues to be optimized, the Seatrec platform has secured backing from investors like Dr. Norge Larson, previous President of Sea-Bird electronics (one of the most successful BlueTech exits in recent years), who led the company’s $2M Seed Round. The startup has also attracted a roster of heavy hitting advisors as well:

Conclusion: How Big Could This Be?

The Blue Economy is a market as large as the ocean is vast:

“The Blue Economy – sustainable, ocean-centric commerce – is expected to generate more than $3T in global economic impact and support more than 40M jobs around the world, according to the Organization for Economic Co-operation and Development (OECD). Aquaculture alone is projected to grow into a $274B per year market by 2025.” - Seatrec Press Release

The TAM for specific startups like Seatrec and Saildrone can be challenging to triangulate, given the variety of use cases their technologies address. But a few relevant data points I dredged up for scale include markets like the spend expected on Oceanographic Monitoring Systems ($2.2B) and Ocean Observation Instruments ($14.8B) by 2030. One can also back into reasonable market-sizes with military and academic budgets for ocean research which reach into the tens of billions as well.

These markets represent an opportunity for entrepreneurs to do good while doing well. The success of startups like Saildrone and Seatrec is integral to accumulating high-fidelity maritime data on a global scale. The treasure troves of data their platforms are collecting help to filter our murky understanding of the world’s oceans and our effects on them.

Cheers,

- 🍋

Subscribe now

Not so long ago, it was magical that we could press a button on our phone and have items miraculously appear at our doorstep. We take it for granted now, but the implementation of two-day delivery by Amazon in 2005 (and same-day delivery around 2009) was a logistical feat. Today, consumers are accustomed to max convenience with max speed– Amazon delivers your package the same day, Uber drops your egg rolls in 30 minutes with 99% accuracy. With the blazing speed of modern logistics and the hyper-competitive retail value chain, where else can efficiencies be rendered?

One possible avenue is last-mile logistics. The final mile is the least efficient, most expensive part of almost any consumer delivery. It’s the stretch in which your economies of scale break down while your labor intensity rises. With cusping advancements in autonomy, there are a number of startups attempting to rethink this link in the supply chain.

Drone delivery, what used to seem like a Jetsonian idea, is rapidly becoming a reality. Startups are attacking this challenge from all variety of angles (ground, air, and sea), but my focus lately has been the fleet of founders zeroing in on electrical Vertical Take Off and Landing (eVTOL) drones to solve last-mile delivery in urban environments. Simply put, delivering your DoorDash via autonomous mini-copter.

Why Last Mile Is Hard

Last-mile logistics, defined broadly as the final step in getting a product from a distribution hub to a customer’s doorstep, has been the bane of many a startup’s existence. Although technology has enabled numerous advancements in connectivity (ordering from your phone, GPS-location of your goods en-route, etc.), the most challenging component for many of these startups is the translation from bits to atoms. In the late teens, a number of founders leveraged the internet’s organizational powers to create a wide range of delivery marketplaces– Uber Eats, DoorDash, Favor– and tackle the byzantine challenges of delivery itself with companies like Fetch, Parcel, and Shyp. The demand clearly exists and the market opportunity is huge, but any service targeting the last-mile itself is an intensely challenging nut to crack. Why is that?

Last-mile delivery can account for up to 53% of supply chain costs, and for good reason. This is where your economies of scale buckle. Where first-mile and middle-mile can capitalize on single-stop (point to point) fixed routes moving bulk product on the backs of planes, trains, and other large vehicles, your final mile often boils down to irregular pathing with highly manual deliveries. Additionally, your primary distribution centers tend to be located just outside of major city centers along well-maintained highway or rail systems. For your final miles, drivers are navigating congested city streets with more variable driving conditions, compounded by multiple stops along any given route.

So, it’s tough, but attractive. The last-mile delivery market is estimated to be over $175B globally, and over $65B in the U.S. alone. Though, that figure should be caveated when considering drone delivery specifically. Aerial drone delivery is around a billion today, expanding 10x by the end of the decade. Capturing even a fraction of the market would represent a massive opportunity, to say nothing of the possibility of induced demand expanding that pie further.

Why Deliver from the Heavens?

Before considering the ‘who’ of drone delivery, I think it’s important to consider the ‘why’. Initially, my gut reaction to aerial drone delivery was as more of a novelty than a serious, scalable, and potentially disruptive delivery platform. I had so many skepticisms— airspace considerations, liability (drones falling out of the sky?), efficiency (single drone + package vs. streetside Amazon trucks), among others. And, while some of those are still valid, I’ve started to think that the juice might just be worth the squeeze.

Fundamentally, it will always come down to cost:

The key is reducing labor costs through scalable drone fleets. Even with the 1:1 drone to package weight limitations, the combined aspects of fleet capabilities and speed of delivery show promising signs of achieving cost parity to legacy delivery channels (cars, vans, etc). And the graphic above simply runs costs today at the legally viable rate of 20 drones per observer, which is the bar that the FAA has allowed for a few startups currently. It’s easy to envision a future where fleets have achieved proven safety standards to scale that number 2-3x, and in fact Wing has been approved for a 50:1 ratio in Melbourne, Australia. Couple this with pathing advantages (‘as the crow flies’ vs. indirect and congested streets) to drive delivery velocity/throughput for the distributor and convenience for the customer, and a more compelling picture begins to emerge.

These cost advantages could all be compounded by network effects as well. If an operator can establish a number of nests across an area, they start to build a network of nodes that can decentralize the delivery chain. This means that, similar to ride sharing, drone operators can shift their swarm resources to the areas with the highest demand. The result could be a more dynamic delivery chain as opposed to hub-and-spoke models which are more rigid.

Time sensitive deliveries are particularly attractive beachheads. On one hand, you have lower ticket, high velocity items like food, where freshness and speed is paramount. Drones can bring both service and sustainability improvements vs. single driver, single package food deliveries that proliferate today (a single drone vs. a thousand pound vehicle ferrying a single burrito..). This market is being attacked by players such as Wing, whom we’ll discuss later, via its pilot with DoorDash. On the other hand, you also have markets with higher ticket, low weight items like blood transportation or temperature controlled pharmaceuticals where speed is paramount, weight is low, ticket price is high, and you can potentially capitalize on fixed routes like pharmaceutical distribution to a hospital. Players like Zipline made their name conquering this beachhead in infrastructurally challenged markets like Rwanda, where the company delivers 75% of Rwanda’s blood supply outside of the country’s capital city.

Drilling down on proximity and weight also helps to envision how broad the urban market can become:

“I mean, if you think about it, 90% of all packages that are delivered into neighborhoods every day weigh eight and half pounds or less. And what’s more interesting- and we’re moving those with 6 to 10 ton trucks- and if you also consider that 90% of those packages sit on a shelf within 5 miles of a customer.. yet that’s not how we’re delivering it.” - Tom Walker, DroneUp (S&D Show)

From my internship in Walmart’s Finance & Strategy Division c. 2017, I distinctly remember leadership discussing the core competency of having a Walmart Store within 10 miles of 90% of the U.S. population, and potentially leveraging these as micro-fulfillment centers. Drones could bring that one step closer to fruition and potentially serve as a key competitive advantage against the Amazons of the world who lack such an extensive physical footprint. If a few parking spaces or an unused rooftop can be transformed into micro drone delivery hubs, that could be a significant driver of adoption and delight for customers:

That being said, this delivery channel isn’t without challenges. By and large, this technology is fairly nascent and has failed to achieve domestic product market fit. Walmart claims that 75% of the DFW population can receive delivery by drone, but the actual utilization is an open question. Frontrunners are certainly growing their deliveries YoY, but the volume is still subscale. A few major challenges include:

• Implementation: Logistically integrating into existing businesses for DTC is challenging. Rooftops of Walmarts are potentially interesting, but other markets like food delivery are open questions: Should they come from drone delivery ghost kitchens? Will they be tacked on to the sides of pharmacies? Or is it self-serve lockers?

  • Zipline’s already way ahead of me with their next-gen platform update - inside, outside, or rooftop (link)

• Airspace + noise complaints: The FAA is still refining its framework for regulating these startups, and there have been challenges demonstrating fleet safety in order to scale observer to drone ratios as well as operations Beyond Visual Line of Sight (BVLOS). There are also questions around restricted zones (critical infra, airports, etc.) and noise complaints in residential areas.

• Consumer behavior: We are in the early innings of user adoption with obvious concerns around safety and package security.

• Competitive alternatives: Autonomous vehicles and ground-based delivery drones have certainly received significant investment, which could bring down costs and render aerial drones less competitive.

Despite these headwinds, there’s still a swarm of players going after this market.

Who’s Building

Zipline - Zipline is one of the clear market leaders, having conducted over a million deliveries across seven countries. The company has amassed a warchest, raising over $800M since inception, and most recently securing a $330M Series F in May 2023 at a $4.2B valuation.

Zipline was originally founded as Romotive, a smartphone-controlled robotics company, by Keller Rinaudo Clifton, Phu Nguyen, and Peter Seid in 2011. The precursor company saw some success, going through Techstars and receiving a $5M Series A from Sequoia, but ultimately pivoted due to constraints around consumer robotics as well as global/societal impact for that original use case:

“The more I thought about it, the harder it was to justify expending my energy on a consumer robotics product that probably wasn’t going to work that well. Robots in the home are a notoriously tough space to crack, and I felt that there were bigger problems I wanted to solve…” - Keller Rinaudo Clifton

This led Rinaudo Cliffton to explore other avenues, eventually piecing together a new founding team to tackle a radically different use case: chronically undersupplied hospitals in rural Africa. In countries like Rwanda, where infrastructure is severely underdeveloped, thousands of preventable deaths occur simply due to a lack of essential supplies– particularly blood. This presented a prime testing ground for aerial drones, coupling the key characteristics above (high ticket, time sensitive, low weight payloads) with a strong willingness to experiment by governments desperate for any solution.

Zipline completed its first delivery in Rwanda in October 2016, and has since scaled its medical supply delivery operations across a number of developing markets including Ghana, Nigeria, and Kenya. The startup has leveraged its first-mover advantage to establish market dominance in places like Rwanda where it delivers an estimated 75% of blood supplied outside the country’s capital city. 

Importantly, these markets provided the team with fertile testing grounds to refine its technology stack (navigation, fleet management, package processing, etc.) and establish its hardware supply chain. Zipline has leveraged these learnings to expand into larger, established markets with more competition. While still focusing on middle-mile medical supply delivery, the company also competes in last-mile logistics and has established partnerships with companies like Panera Bread and Walmart in the U.S. Today, Zipline has completed over a million deliveries worldwide.

An aside: it’s totally worth checking out their bad ass delivery system in Rwanda, including an aircraft carrier-like sling shot + grab wire system for launch/landing.

Wing - Wing has a slightly different origin story. The company was started in 2012 as a project within Alphabet’s X (f/k/a Google X). Wing originally explored the delivery of medical supplies similar to Zipline, but made a pivot to broader consumer delivery early on. The startup completed its first deliveries in 2014, and eventually graduated from Google X in 2018 to become an independent subsidiary beneath Alphabet. For the duration of its life, Wing was built by James Burgess and Adam Woodworth, who served as CEO and CTO, respectively. Burgess stepped back in early 2022 for health reasons, with Woodworth remaining at the helm today.

It’s hard to pinpoint precisely how much has gone into Wing, given its subsidiary status. The company sits beneath Alphabet’s ‘Other Bets” line item, which discloses $1.2B in revenue with operating losses of $4.5B, but this bucket also includes projects like Waymo, Verily, and others. That being said, they have offered a few rough figures for commercial traction. In 2023, Wing had completed approximately 350K deliveries (vs. Zipline’s 1M commercial deliveries). This appears to be accelerating significantly though:

“On average this summer, we launched a new drone facility in the DFW area every two weeks. And we’re planning an even faster pace of expansion for the rest of this year and into 2025” - Wing Blog, Oct. 3, 2024

Given the similar timelines from founding through commercialization, it’s interesting to examine why Wing may not have achieved quite the same scale as Zipline yet. Two thoughts— strategy and funding structure. From a strategy perspective, going after consumer deliveries may have drawn more regulatory scrutiny and slowed Wing’s overall time to market vs. Zipline’s high pain point, low regulatory environments like Africa where drone delivery solved a clear and obvious problem. Drone deliveries in countries with challenged infrastructure represent more of a stepwise improvement vs. marginal efficiency gains. Then on the capitalization front, there is little reported for how Google X allocates resources, but one could speculate that being a wholly owned subsidiary shielded the company from market/investor pressures that either A) may have demanded more rapid results or B) provided the virtuous cycle of a deluge of dollars following early traction to accelerate far more rapidly. These are broad speculations on my part.

Another interesting strategic decision is the development of Wing’s OpenSky app. Today, the app helps operators map routes, navigate regulations, and receive approvals for controlled air spaces. But this could potentially be a bet on the proliferation of drones, staking an eventual market position as the base protocol through which myriad independent drone operators will run their machines to coordinate routing. Here’s a great promo video from a drone photographer using OpenSky in his work.

Incumbents - Of course, this space hasn’t gone unnoticed by the incumbents. Amazon, DHL, FedEx and others all continue to explore last-mile drone delivery solutions of their own. Amazon recently scored BVLOS approval for its MK30 drone, and is rolling out drone deliveries in markets around Arizona. Alternatively, Walmart has historically favored partnering with specialized delivery companies, working with Wing, Zipline, and even investing in DroneUp.

Conclusion

It sounds crazy. Crazy enough it just might work. People have been toying with the idea of drone delivery for decades now. Bezos himself (optimistically) predicted in 2013 that drone deliveries could be as close as a couple years out as Amazon began experimenting with the technology.

Today, I think there are a confluence of factors accelerating potential adoption. Moore’s Law has propelled improvements in AI and computing power, boosting drone technology. This has worked in conjunction with the early movers like Zipline beginning to achieve manufacturing scale to march further down the cost curve. But what has really been a black swan accelerant for the drone industry has been the deterioration in global geopolitics (Ukraine, Israel-Palestine, etc.). Ukraine specifically has established itself as a proving ground for the future of warfare, and pushed drone performance and swarm capabilities by leaps and bounds. This is to say nothing of the potential for a near-peer conflict in the South Seas, which of course has also accelerated logistical development within the military innovation community as well.

We’re still a long way from broader adoption, but what once seemed a flight of fancy might be closer than we think.

- 🍋

Subscribe now

There’s a meme in tech culture about hiring ‘10x engineers’. It’s the idea that a single exceptional engineer doesn’t merely add value to a startup in a linear (+1) fashion, but that the right talent can exponentially improve a product. And it’s an acknowledgement that these hires aren’t evenly distributed.

I recently found myself extrapolating this idea to the relationships in my life:

Relationships follow power laws. You will have 4 or 5 in your life that profoundly impact you, with the rest trailing off dramatically. So, if the impact of relationships fails to disperse across 1x, 2x, 3x improvements in our wellbeing, but are far more pronounced in actuality (1x, 10x, 100x)... Why is it that we don’t conduct ourselves accordingly?

One answer is that it’s easy to be complacent, especially with our time. The things we’re given for free, we often fail to value. And this counts doubly for the relationships that feel effortless. Precisely why we fail to miss people until they’re gone.

We all have the friends that are fireworks in our life- seen once a year, dazzling in their briefness, and magical in their effects. Yet the scarcity of these encounters are often self-inflicted. We have the time, we have the ability, but we fail to invest accordingly. We enjoy these precious little human nuggets when we see them, but we otherwise allow life to intervene. 

Moreover, we allow life to be leeched away by menial tasks, lackluster dinners, and draining conversations. It’s like the scope creep of the meaningless bullshit in our lives. You postpone a couple lunches, and a year slips away since you’ve seen your best friend.

It’s subtle, and it’s challenging. But this idea is worth remembering. As relationships are all that we ultimately have, to neglect the most impactful is tantamount to pissing away our precious time.

This sentiment isn’t uncommon-  “20 years from now, the only people that will remember you worked late are your kids”. But it should not be forgotten by those of us who haven’t quite reached that stage yet. With each advancing year, the responsibilities add up. In the present moment, you are likely the most healthy, the most flexible, and the least burdened that you will henceforth be. You think your knees hurt now? Just wait.

And so, I suppose the first step is recognizing that you have a problem. Half the battle is recognizing the value of the precious resource we have (time), and considering exactly how the hours are being spent: invested or squandered. Plan the trip, take the lunch, and make the call. It’s always worth the investment to double down on those who care.

- 🍋