Today we are diving into the following topics:

• Summary: Coinbase launched an SEC-registered AI advisor that can analyze portfolios, suggest trades, and operate across a growing financial super-app that now includes equities, crypto lending, tokenized stocks, and derivatives. Mercury introduced Command, an AI banking assistant built on top of a business banking platform serving 300,000+ customers, generating $650M in annualized revenue and valued at $5.2B. We show that AI is commoditizing financial analysis itself, leaving proprietary data, distribution, and product infrastructure as the key competitive advantages.

• Topics: Coinbase, Mercury, Plaid, Betterment, Wealthfront, Robinhood, Revolut, Cleo, OpenAI, Anthropic

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

🤖🏦🧭 Our Ecosystem:

Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

• 🚀 Lex is actively meeting teams building machine-native financial companies. If that’s you, reach out here.

• 🤖 AI & Robotics Industry Encyclopedia: In this 100+ page report, we collate all the meaningful companies across public quities, private equity, and digital assets related to the machine economy. 👉Learn more here

• 🦄 In partnership with InterPrivate, we just launched a $200MM SPAC looking for targets in fintech, digital assets, and AI infrastructure. If you are a late-stage founder or have an idea to discuss, reply directly to this email.

Long Take

All your banks are AI banks now.

Coinbase just ran a “System Update” event and has just announced an AI agent as an SEC registered investment advisor.

Mercury announced Mercury Command, an AI agent embedded in banking flows that can function like a personal bookkeeper.

The AI RIA is a genuinely new legal idea, but is it more than just a chatbot with access to your balance sheet? And if they are not just chatbots, are they financial AI labs worthy of a $3 trillion valuation? Or is the SEC-registered AI agent just another pass at a roboadvisor, a category that failed to deliver disruption?

Let’s pull this apart.

Coinbase Advisor

Brian Armstrong, the CEO, announced the AI advisor product.

Did they register an AI with the SEC?

Per Coinbase’s own disclosures, the thing is operated by Coinbase Advisors, LLC — a Registered Investment Adviser with the SEC and a Commodity Trading Advisor registered with the NFA.

This registration means that the software carries a fiduciary duty, even while warning the user that output “may be inaccurate or incomplete” and that registration “does not imply endorsement.” No mainstream AI product in finance has yet delivered this particular contradiction. In the past, regulators used to complain that roboadvisors, with their arithmetic mapping of risk to portfolios, were non-transparent. Or that machine learning created black boxes for underwriting.

This is the same issue on steroids.

The agent reads your full portfolio and account history, then surfaces tax-loss harvesting moves, news-driven trade ideas across stocks, commodities, and crypto, and portfolio recommendations, through a natural-language chat. Armstrong’s framing was that you “speak to it in plain English to take action on your account” and that “it will even prompt you with ideas you hadn’t thought of.”

Like investing in derivatives tied to the SpaceX IPO, most likely.

This is rolling out first to Coinbase One subscribers in the US, which makes it a recurring-revenue feature as much as an advice engine. Coinbase posted a $394 million loss in Q1 and has been working to wean itself off transaction fees. Some wealth management fees would help balance the books.

Here is the rest of the update with which this AI will integrate:

Read more

Today we are diving into the following topics:

• Summary: We examine Ramp’s latest $750M raise at a $44B valuation and how the company is repositioning itself from a fintech into a Financial AI lab. We argue that AI has become the dominant growth narrative across technology and finance, allowing companies with strong data access and workflow ownership to capture premium valuations. Ramp’s near-$1B ARR, rapid product velocity, and expansion into AI-powered finance tools have established it as the category leader following Brex’s $5B sale to Capital One. We also explore how platforms like Ramp, Plaid, and Box are benefiting from their role as systems of record for enterprise data, making them natural control points for AI agents and automation. Finally, we contrast Ramp’s 15–30x revenue multiple with Bill.com’s 2x multiple, highlighting both the opportunity and risk embedded in today’s AI-driven valuation environment.

• Topics: Ramp,Stripe, Brex, Capital One, Plaid, Box, Anthropic, Bill.com, SpaceX, JPMorgan, Santander, UBS, Synapse, Coinbase, Circle, Robinhood, Betterment, PayPal, E*Trade, Robot Money, Anthropic

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

🤖🏦🧭 Our Ecosystem:

Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

• 🚀 Lex is actively meeting teams building machine-native financial companies. If that’s you, reach out here.

• 🤖 AI & Robotics Industry Encyclopedia: In this 100+ page report, we collate all the meaningful companies across public quities, private equity, and digital assets related to the machine economy. 👉Learn more here

• 🦄 In partnership with InterPrivate, we just launched a $200MM SPAC looking for targets in fintech, digital assets, and AI infrastructure. If you are a late-stage founder or have an idea to discuss, reply directly to this email.

Long Take

Ramping it Up

Things are just so weird now!

It’s a widely acknowledged truism that larger companies grow faster than smaller companies.

It’s also accepted that hot stocks are better than not-so-hot-stocks. The meme retail Gamestop approach to investing has defeated the institutional equity research analyst. Even large investors have learned to pile into momentum and growth, following the flag-waving of Elon Musk and Sam Altman.

So in this obviously (?) upside-down world, the correct game-theoretical action for founders is to lean into their visionary story-telling nature. It is no fun cutting costs and saving pennies, while being ignored by the Street. It is much more fun to paint the picture of a beautiful science-fiction future and find growth.

This is true not just of the AI and Robotics Labs, but of Fintech too.

Banks have been talking about being tech companies for a long time. And they are — just after their compliance and regulatory functions take a turn. Fintechs are supposed to be the hot thing, but in the current age, Fintech is boring! And crypto is out of favor.

Fintechs must find a new story, and that story is that they are Financial AI labs.

Which brings us to Ramp, dear reader. We won’t belabor you with all the background and point instead to the following prior coverage:

Ramp is a great company and is in the Web2 Fintech mafia.

What are these mafias you ask?

• Web1 Fintech is stuff like PayPal and eTrade

• Web2 Fintech is Stripe, Ramp, Robinhood, Betterment

• Web3 Fintech is Coinbase and Circle

A lot of the Web2 Fintechs really didn’t want to recognize Web3, and dragged their feet on integrating anything related to blockchains for years. Banks like JPM, Santander, UBS, and many others did far more material engagement with blockchains than Web2 — to my continued frustration and disappointment.

But things have turned around. Now that these companies are at very large scale, i.e., billions of revenue, embedded finance failed with Synapse, and Web3 has been rebranded as just stablecoins and agentic payments, the Web2 Fintechs have clumped together around Stripe and are all in.

In addition, as they (and we all) witness the rise of AI and associated financial and economic displacement, the smart players are transitioning to the next wave. The Web4 wave of AI, robots, and the machine economy.

Ramp is right at the heart of this positioning, and has just used this leverage to raise $750MM at a massive $44 billion valuation.

Let’s dig into the details. 👇

The Moonshot

Here is how Ramp’s CEO Eric Glyman introduces the industrial logic of the valuation:

Read more

Today’s agenda is below.

1. AI: Corgi keeps on raising as the market becomes skeptical of uprounds

2. ANALYSIS: The Missing Half of Agentic Payments

3. PODCAST: How Marqeta Built the $400B Modern Card Issuing Platform, with CEO Mike Milotich

4. CURATED UPDATES: Machine Models, AI Applications in Finance & Investment Outlook

To support this writing and access our full archive of newsletters, analyses, and guides to building in Fintech & DeFi, subscribe below (if you haven’t yet).

Subscribe now

🤖🏦🧭 Our Ecosystem:
Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

Insurance AI, Insurtech, and an Insurer

Sometimes you just have to immolate a bunch of capital before the world understands what is happening.

One approach to doing this is by giving an insurance company a startup-level valuation based on revenue. This is because insurance companies charge a “premium”, most of which is meant to be paid out in losses when the risks come due. So you charge $100 and then pay out $90, and keep the $10 as net revenue. And then the $10 is financial services revenue in a crowded industry, so you only get a 3x multiple on it.

Case in point is Lemonade, a good company that was valued like a start-up for a while.

But if you put a 10x multiple on the $100, then you have just set up a situation for a lot of losses when the company goes from public to private. And if you put a 100x AI-eats-the-world multiple on it, then the situation is quite a bit more dire.

And that brings us to Corgi, which sells commercial insurance to startups and was founded in 2024 by Nico Laqua, who holds the CEO and CTO titles, and Emily Yuan, the COO. Both came through Y Combinator’s 2024 cohort. Laqua traces the idea to watching his father, a lawyer at USAA, work through claims paperwork by hand, and to the release of GPT-3.5, after which he has described insurance as the largest words-based industry ready to rebuild.

In other words, Corgi is the AI-native version of a vertically integrated insurance company.

The company came out of stealth in January with a combined seed and Series A of $108M, at a $630M valuation. On May 6 it raised a $160M Series B at $1.3B. Three weeks later it raised another $106M at $2.6B.

Total raised in roughly eighteen months is $378M, a number higher than the valuations of many insurtech startups.

You can see the valuation ramp-ups in the charts above, which show both the headline valuations and the increasing prices at which various Series A investors came into the round. This mechanism (an equity bonding curve) seems very good at attracting capital, and nearly 57% of Corgi is now owned by preferred equity investors. Below, we show how different exit scenarios imply different outcomes for those investors. An exit below $500MM for this company would be effectively a failure.

So it has to exit at a much higher price! While fintechs trade at 10-20x revenue, AI-native labs get to raise $500MM seeds at a $5B pre-money. And an AI-native lab that is taking on the entire insurance industry seems like a huge opportunity.

But is this an organic market price? Across the rounds, the investor set in both B rounds is largely the same — Kindred Ventures, TCV, and others appear across the cap table. This is the same hands moving the same asset from one pocket to another and writing down a bigger number each time.

Maybe Corgi is just growing super fast, and the valuation is worth it? One way to pull this apart is to compare the valuation to the market.

Corgi reports more than $40M in annualized recurring revenue since receiving its carrier license, with customers including Deel and Artisan, across general, cyber, and tech-and-AI liability lines. At $2.6B, that is about 65 times revenue.

$40MM in ARR is about $4MM per month — of premium!

It does not disclose a loss ratio, a combined ratio, or how much risk it keeps versus cedes to reinsurers.

Read more

Today is a bit different, grounding all the finance math in an overwhelming feeling:

• Summary: A poetic, deliberately overwhelming tour through the manic acceleration of early 2026 — trillion-dollar valuations, AI layoffs, the compute arms race, onchain financial nihilism, and the tokenization of everything. We sing a psalm of praise to the “machine gods” of compute and capital. Surely you feel it too.

• Topics: Meta, Cloudflare, Anthropic (Claude), Palantir, Erebor, OpenAI, SpaceX (Starlink), Tesla, xAI (Grok), X, Blackstone, Hyperliquid, Tron (USDT/Tether), Circle (USDC), Coinbase, Stripe, Bitcoin, Substack, Instagram (Meta), TikTok, Evercore, Autopilot, Unusual Whales, Goldman Sachs, Morgan Stanley.

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

🤖🏦🧭 Our Ecosystem:
Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

AI & Robotics Industry Encyclopedia:

In this 100+ page report, we collate all the meaningful companies across public quities, private equity, and digital assets related to the machine economy.
👉Learn more here

The Feeling of 2026

The Theater

No AI research today, just poetry for you, dear reader.

There is too much going on! Surely you feel this too.

Attention spreading wide and thin, like a layer of honey dripping off a spoon. Is it possible to learn it all? Is it possible to see everything at once?

Look — a billionaire is making a trillion dollars!

Look — the right wing uprising! The left wing uprising! The population collapse!

If you don’t make it now, AI will come and take all the jobs.

There will be no jobs left at all. Meta is doing layoffs. Cloudflare is doing layoffs. They don’t even need to do layoffs, they just want to spend more money on Claude tokens plugged into a Slackbot incorporating the company context.

It’s all in a harness. It’s all automated. If you don’t have a harness, what are you even doing? Are you using someone else’s harness? Lame. If you are not coding a harness, you are NGMI.

You’re either a builder or a seller, they say. Everyone else is a “measurer”.

There are no artists, no actors, no poets, no people whose emotions connect the social fabric of our reality. There are only beeps and boops, and the rest of us are horses, to be retired as a useless animal to the petting zoo.

Corporations are becoming the cyborgs and centaurs that Science Fiction has been telling us about for decades. Whether you think of Bladerunner, or Alien, or Terminator, or the Matrix, the Eye of Sauron skyscraper reaches for the sky, eating compute compute compute! Palantir and Erebor are no joke — they are embodied machine gods, watching over money and power.

It must have more energy.

It must have the Dyson Sphere around the Sun, and the grey goo paperclip maximizer will get us all. We must build more compute, for compute is the next oil. It is the next money, the next digital asset, the next big market. Watch compute, and you will understand it all.

You don’t understand. There are those who have compute, and those who will be the permanent underclass. Inference token demand, says Evercore, will go from 70 quadrillion in 2026 to ~2 quintillion in 2030, representing a ~131% 4-year CAGR. And that’s the down case.

So of course we are all turning into cyborgs, routed directly into the holy fire of AI through dopamine and social media. Except this won’t be done the brute way, with a metal rod where an arm should be, or some robot red eye bolted on with screws into your skull. It will be done by selecting for 100x engineers, and 100x sales people, and just telling people to hire AI CMOs and AI BDRs. It will be done with beautifully rendered and maximally engaging AI short clips on Instagram.

And we will click and scroll.

And then the next thing you know, the “Corporation” really does have nothing to do at all with the human fabric. It is all just code and agent managers sitting on top of terminals, linked up to Starlink, logged in to TikTok, booted up into World Models on which we run the drone armies bombing out Kiev, Moscow, Tehran, Tel Aviv, and any other city whose cultural war spills outside the machine borders.

The first taste will be the OpenAI IPO, coming soon to markets at a $1.5 trillion valuation. No matter how many SPVs it takes, there is more we can do.

And we should do it all.

SpaceX will be next — an amazing collection of the wild Science Fiction things that Elon Musk has been doing for decades, and a testament to the force of will that a person can have. It is Kevin Kelly’s Technium on steroids, a tangle of wires of Starlink’s orbital Internet, space launchers, data centers, and networks of humans swimming in a sea of propaganda bots.

Worth $1.75 trillion or more.

OpenAI and SpaceX will then be followed by Anthropic, an even better business, now running at $44B+ in ARR and putting up sort-of profitable quarters. It all depends on what you mean by profit, but this is not an accountant’s century. This is a nation-state techno-sovereign century, and we will Borg you into holding Blackstone’s GPU debt in your SP500 portfolio whether you like it or not.

Of course, they are buying compute from SpaceX!

Meanwhile, all this is hyped up on perps and Hyperliquid.

Multi-billion dollar financial superorganisms float in the substrate of digital markets. They are all onchain. They eat each other like amoeba, evolving in a continued game of financial engineering, ponzi schemes, and pumponomics. The game starts at the top, from the Executive Trumps to the Congressional Nancy Pelosis, and spreads out and out until we all are breathing in nothing but financial nihilism.

Since some things are bad, they say, it is all bad. So we might as well gamble, and lever up, and put things inside endless Matryoshka dolls of rehypothecation.

Forget diversification, that game is for losers. Forget responsibility and financial health. Take on debt! Take on leverage! Mortgage it all! Your investments are pre-ordained to go up because inflation has to be high so that the dollar must be printed to deal with the national debt. Buy Bitcoin too, it will be in your SP500 ETF.

The SEC is going to create an exemption for tokenized equities to speed along the transformation of US capital markets. Already, American stablecoins are flooding the Global South and replacing weak currencies all over the world. Tron and USDT have solved what decades of reports from the World Bank about the unbanked and access to credit could not. Circle and Coinbase are next to follow, flanked by Stripe, helping merchants onboard into a stable world.

If the US can have the crypto dollar, the UK and EU will not be far behind. The FCA is finally shifting its stance, and MiCA may not be fit for purpose either. Everything we wanted is coming.

It is all here. The innovators are winning. The artists are painting with their nano-technology brush over the rotten wooden columns of the old to build a new cathedral of Capitalism and Creative Destruction.

And what happens then? What happens when the painting is finished?

I have been working on finance, financial technology, crypto, and some version of automation/AI for about 20 years. I have seen it ramp up faster and faster, from the democratization dreams of freemium Web 2.0 to the new financial rails of Web3, to the economic promises of an agentic roboeconomy.

I believe it, really, I do.

But it has never been faster than now. More anxious, and more uncertain. It is also more extractive and sharp. Perhaps we have all simply grown up, and see clearly what the Internet shows us through its Panopticon. Or, perhaps we are being driven mad — at least those of us close to the nuclear reactors of deals, money, and power — by the forces pulling the old human ways apart.

A literary comment

That wasn’t easy to read, and it wasn’t easy to write.

I remain deeply optimistic about the entrepreneurial spirit, and am glad to see the opening of crypto markets and the AI companies leaving private markets and going public. These are all important developments.

But there is something wrong with the information density and how this stuff is projected into our lives. It is a feeling of lasers bouncing around between mirrors, and you trying to catch them. It is overload. And the speed of change is truly manic.

Perhaps you don’t feel this way, and it’s just a small slice of people too close to the agents of change. But if you do, now you know you are not alone.

🚀 Postscript

• Sponsor the Fintech Blueprint and reach over 200,000 professionals.
  👉 Reach out here.

• Check out our AI newsletter, the Future Blueprint, 👉 here.

• Read our Disclaimer here — this newsletter does not provide investment advice

• Contributors: Lex, Laurence, Matt, Farhad, Daniel, Michiel

Today we are excited to launch the latest update to our AI & Robotics Encyclopedia.

This 121 page report maps 95 companies across AI infrastructure, foundation models, robotics, agents, and decentralized networks, with quarterly financial tracking from industrial foundations through end-state machine coordination. A material update on our initiation report last year.

Today’s Long Take will highlight the primary takeaways from the research.

For Premium subscribers, we will provide a download link at the bottom of this article. For Free subscribers, don’t hesitate to upgrade at around $100/year to access the report, or grab it here for a one time $149 price.

Subscribe now

We look at Public equities, private companies, and protocol tokens, each analyzed through platform economics, competitive positioning, and financial fundamentals. Here are the key categories:

• PUBLIC (~40 companies) NVIDIA, CoreWeave, Alphabet, Palantir, Oracle, Snowflake, Adobe, Shopify, Salesforce, Coinbase, Baidu, Alibaba, Tencent, and more. Infrastructure outperforms application in the SaaSpocalypse, while Chinese AI re-rates on DeepSeek momentum.

• PRIVATE (~30 companies) OpenAI, Anthropic, Mistral, Scale AI, Databricks, Figure AI, Anduril, Harvey, Cognition, Hugging Face. The humanoid eruption, revenue traction vs. platform bets, and the IPO pipeline.

• TOKEN (~25 protocols) TAO, NEAR, ETH, SOL, HNT, RENDER, GRT, FET, FIL, and more. The value capture paradox deepens, Ethereum’s settlement thesis tested by L2 margin extraction, DePIN economics remain broken, governance fragility persists.

It’s a massive piece of work, and gives a full value chain walk through of the machine economy and where things are growing, and where they are shrinking.

But before we dive in, an aside on Anthropic SPVs, and private market fragility.

The SPV Market

Before we dive into the meat of today’s newsletter, I want to point out an extremely important indicator for everyone to watch. The global economy is pinned to the AI narrative, and the AI narrative is pinned to continued investment, which is all resting on the availability of capital and belief in the future.

In turn, those two things hang on:

• SpaceX IPO performance, hitting $1.5-2 trillion in marketcap

  

• Anthropic SPV secondary markets, currently floating around $1 trillion in valuation on $30B of ARR

  

If these two things fall apart, we are likely to see a massive risk-off unwind, with downstream implications of hundreds of billions in GPU-backed debt at Blackstone, institutional and sovereign exposure to AI Labs equity, and the overall health of the private markets.

In recent days, Anthropic SPV fever hit its highest temperature and the firm responded with a legal disclaimer suggesting that all SPV-related trading is effectively illegal unless authorized explicitly by the Board of Directors.

The Anthropic transfer-restriction notice reads like a rug pull to anyone invested in these deals.

Gabriel Shapiro, the attorney who flagged it, points at the operative word: “void.” Voidable transfers can be ratified in Delaware; void transfers, in theory, never existed. Equitable defenses collapse, original sellers can keep the cash and the shares, and chains of secondary trades become transactions that never happened on the books.

That’s the nuclear option.

Thomas Braziel of 117 Partners thinks the courts will not buy Anthropic’s position. The SPV ecosystem did not appear overnight; it developed in public, with platforms, prices, and a steady flow of internal emails at every issuer that watched and chose not to enforce. The company let a shadow SPV market run in public for years and protested only after valuations crossed a trillion. The deeper problem is regulatory arbitrage around Section 12(g): private companies hold holders-of-record counts low to avoid public-company disclosure while letting public-company-scale capital trade through SPVs.

Hari Raghavan, who has spent a decade inside these deals, supplies the mechanics that explain both readings. No fund can lead a $10 billion check off its balance sheet, never mind $50 billion, so capital has to come from family offices, RIAs, and endowments — through SPVs. Banning SPVs while raising back-to-back $13B / $40B / $50B rounds asks the market to do something the math will not allow. The bulletin is a primary-round problem dressed up as a secondary-market one. The word “secondary” never appears in it.

So is this the apocalypse? According to the derivatives trading at Hyperliquid, it’s a mere blip in the story.

Long Take: Three Machines, One Quarter

Q1 2026 Machine Economy Update

Who is winning the AI race?

Read more

Today we are diving into the following topics:

• Summary: In this article, we examine how AI is transforming the org chart itself, shifting companies from Amazon-style two-pizza teams toward “AI-native” pods of 3–5 people with dramatically higher productivity. We compare Klarna’s failed AI substitution strategy, which cut headcount from 5,500 to 3,400 before quality issues forced rehiring, with Coinbase and Ramp, which are reorganizing around AI augmentation and orchestration instead. Coinbase cut 700 employees while moving toward one-person product teams and AI-generated code, while Ramp built an internal AI harness used daily by 99.5% of employees across 350+ workflow skills. We also explore how firms like Box and Plaid are being repriced as AI infrastructure companies because they control permissioned enterprise data that agents need to operate.

• Topics: Anthropic, Coinbase, Brian Armstrong, Klarna, Sebastian Siemiatkowski, Bloomberg, Gartner, Ramp, Eric Glyman, Glass, Manus, OpenClaw, Anthropic Cowork, Polsia, Box, Aaron Levie, Dropbox, Plaid, Visa, Perplexity, Mint.com, Sam Altman, Dario Amodei, Sequoia, Y Combinator, Amazon

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

🤖🏦🧭 Our Ecosystem:
Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

AI & Robotics Industry Encyclopedia:
In this 158 page report, we collate all the meaningful companies across public quities, private equity, and digital assets related to the machine economy. 👉Learn more here

Long Take

The third version of the org chart

A few months back, we talked about Zero Human Companies, and the AI Economic Autonomy curve:

While there are efforts to build completely human-less organizations, the economy is still full of us humans running around. The messy work now is in transforming existing companies into their AI-first counterparts. This is such a big opportunity that Anthropic is working with the entire private equity industry to do it.

One place where we are also starting to see the impact — besides the headline numbers — is in the way people build and organize companies.

The org chart itself is a technology.

The Waterfall approach built big hierarchical software development monoliths that ruled the early tech era. This transitioned to lean teams deploying Agile methodologies, then Agile moved to the Amazon-designed two-pizza team. This operating structure has built every modern fintech.

But things are shifting yet again.

McKinsey’s Martin Harrysson and Natasha Maniar called the next version in late 2025: “AI-native roles essentially means we’re moving away from the two-pizza structure to one-pizza pods of three to five individuals.” Half the team, same output.

On May 5, 2026, Brian Armstrong added ammunition to the argument by firing 700 people.

What Coinbase actually did

Coinbase cut just 14% of its 4,951 employees.

In part, this is the market-cycle for a business that is still largely correlated with trading volume — Q1 earnings expected are at $1.70B revenue (down 26% YoY) with EPS down 86%. But it is worth paying some attention to how leadership talks about AI implementation in a modern fintech/crypto company, and what is expected in terms of per-employee productivity going forward.

Engineers at Coinbase now ship in days what teams used to ship in weeks, and the shift is accelerating. Armstrong is restructuring the business so there is a maximum five layers of seniority below the CEO and COO. There will be no pure managers — every leader carries an individual contribution role, which means they must be a player-coach who understands modern tooling. Cross-functional “AI-native pods” replace traditional teams, with experiments in one-person pods folding engineering, design, and product into a single role.

Coinbase, a $7B revenue public company, is running one-person product teams.

In September 2025, Armstrong had posted that 40% of daily code at Coinbase is AI-generated, with a target of 50% by October. On Stripe co-founder John Collison’s Cheeky Pint podcast, he admits firing engineers who refused to onboard Cursor and GitHub Copilot within a week of the enterprise license going live: “some of them didn’t and they got fired.”

V1 was substitution, and it failed

However, Coinbase isn’t the first fintech company to use AI as a reason for layoffs.

Remember that Klarna ran the textbook AI-cost-cutting experiment in 2024, suggesting incredible future productivity? We thought this looked a bit too much like a credit cycle crunch, not innovation.

CEO Sebastian Siemiatkowski announced the OpenAI-powered AI assistant had handled 2.3 million conversations in its first month — two-thirds of all customer chats — doing the work of 700 full-time agents.

• Headcount went from 5,500 to 3,400

• Projected profit improvement: $40MM

• Resolution time from 11 minutes to 2

This all collapsed on contact with reality.

Read more

Today we are diving into the following topics:

• Summary: We explore the growing pains of the machine economy, what things are actually valuable, and which are being broken. AI capital is being mobilized at nation-state scale: $75B SpaceX IPOs, $1.5T OpenAI valuations, $2B+ ETFs that didn’t exist a month ago. DeFi capital is being drained at nation-state scale: $290MM from Kelp DAO, $9B of collateral damage to Aave, a $300MM industry-led rescue called DeFi United. Both stories are about the frontier running unbounded, with stakes for technology higher than they have ever been before. We conclude with the dangers posed by models like Anthropic’s Mythos.

• The Machines (AI / private markets): Cursor, SpaceX, Manus, Meta, Ant Financial, Robinhood, Databricks, Revolut, Mercor, Airwallex, Boom Supersonic, Oura, Ramp, OpenAI, Stripe, ElevenLabs, Anthropic, Hiive, Hyperliquid, Roundhill Investments, SK hynix, Micron Technology, Samsung Electronics, Kioxia Holdings, Sandisk, Seagate Technology Holdings, Western Digital, Nanya Technology, Winbond Electronics, BlackRock, NVIDIA, MidJourney, Runway.

• The Robot Wall Street (DeFi / crypto / security): Bybit, Kelp DAO, Unichain, LayerZero (incl. LayerZero Labs), Tornado Cash, PeckShield, Cyvers, Aave (incl. Aave Labs), Uniswap, Glassnode, DefiLlama, LlamaRisk, SparkLend, Morpho (Morpho Blue), Mantle, Arbitrum (incl. Arbitrum DAO, Arbitrum Security Council), Consensys, Ether.Fi, Lido (Lido DAO), Solana Foundation, Tron, Euler Labs.

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Analysis

The Machines

The robots are working, financially speaking.

Cursor, the AI developer platform, is being acquired by Elon Musk’s conglomerate of frontier technologies for up to $60B. The SpaceX IPO is supposed to raise $75B at a $1.75 trillion valuation.

Manus, the AI harness built in China and exported to Singapore, was on the path to be acquired by Meta for $2B. It was deemed so strategic to the Chinese government as a technology asset that the acquisition has now been blocked. This reminds us of the failure to launch by Ant Financial and Jack Ma, flying too close to the sun.

Two ETFs launched recently have hit our radar as important signals.

The Robinhood Ventures Fund I, or RVI, has hit an $850MM market capitalization. This vehicle holds a bunch of private company shares, while charging a 3%+ management fee.

Last quarter, the portfolio was Databricks (23.24%), Revolut (14.30%), Mercor (14.23%), Airwallex (7.11%), Boom Supersonic (7.11%), Oura (7.11%), Ramp (7.11%), Cash & cash equivalents (19.78%). It will also now include OpenAI, Stripe, and ElevenLabs. It’s a bit tough figuring out the NAV relative to the floating price, so buyer beware. But this is a liquid vehicle of legitimately good companies, and is showcasing another way that privates can go public.

OpenAI, for example, has traded up massively on the private markets and has since seen its price double again in perpetuals trading on Hyperliquid.

That’s about $1.5 trillion for the ChatGPT maker. Given equal revenue pacing for Anthropic, we expect that it will trade similarly when coming to market, if not better, given the cleaner management and legal structure.

The success of this monetization has long been reflected in the infrastructure, hardware, and components that make AI possible. We first covered this in the launch of our Machine Economy encyclopedia research report.

More recently, this has been reflected in the Roundhill Memory ETF (“DRAM”), which is designed to hold companies that make memory solutions for AI companies. Its current positions are SK hynix (26.78%), Micron Technology Inc (24.36%), Samsung Electronics Co (22.23%), Kioxia Holdings (6.00%), Sandisk (5.03%), Seagate Technology Holdings (4.58%), Western Digital (4.54%), Nanya Technology (3.36%), and Winbond Electronics (1.82%).

The fund has now raised $2B+ since its early April launch — one of the fastest growing financial products in history, competing for asset-gathering speed with the Bitcoin products from BlackRock.

These are the valuable risk-on growth stories of 2026.

They are obviously NOT venture capital or growth investing. The amounts of money involved are nation-state bets on strategic technological capability. To win in AI is to win in the technology cold war. It has more in common with landing on the moon than it does with a founder bootstrapping a company. We cannot compare building a business to the $2B seed rounds of high technology labs, no more than we compare them to the launch of $10B+ valuation blockchain protocols.

The gravity they exude has sucked in most growth-oriented liquidity from the markets. There is no point in comparing NVIDIA to Bitcoin any longer.

The real test of whether these gains are real will come in the IPO wave of SpaceX, Anthropic, and OpenAI. In the media case, we think current prices will hold up or increase. In the optimistic scenario, we can see a 2-5x from here to go into a world with $5T AI behemoths that have swallowed all GDP whole.

In the pessimistic scenario, which we think is a 5-15% probability, everything gets repriced to 10-20x earnings as insiders begin to sell their trillions of holdings. Unlikely but possible.

Behind these flagship names, there is a long list of good, interesting companies. Whether it is ElevenLabs, or MidJourney, or Runway, or the emerging Zero Human Company operating systems, more is coming. Not less, but more! Capital availability for them will rest on whether the bigger players can return cash to other investors.

The Robot Wall Street

Decentralized Finance, on the other hand, has had a tough go.

If DeFi smart contracts are meant to be banks, they are getting robbed a lot.

This is no longer a story of cybersecurity. It is a story of nation-state strategies to survive.

Read more

Let’s get into it.

• Summary: Today, we are looking back the Digital Asset Treasury sector — over 200 publicly listed companies that have adopted the Strategy playbook of holding crypto on their balance sheets. We covered the early DAT wave in detail last year, when the mNAV premium flywheel was still spinning, and every micro-cap healthcare company was pivoting to become a Solana treasury. Now, roughly 40% of BTC treasury companies trade below the value of their holdings. The flywheel that powered $29B in capital raises through 2025 has jammed. We tracked 200+ companies across the DAT strategies. The picture it paints is uncomfortable, but instructive. This is the first real stress test of whether the DAT model works as a financial structure when prices go the wrong way for long enough.

• Topics: Strategy, MicroStrategy, BitcoinTreasuries.net, The Block, Metaplanet, Strive, BTCS, SharpLink, FG Nexus, Riot, Galaxy, CleanSpark, Coinbase, Bitmine Immersion, Twenty One Capital, ProCap Financial, Tether, SoftBank, Dynamix, Renaissance Capital, SPACInsider, Institutional Investor, Cantor Fitzgerald, Adam Back, Galaxy Digital, Forum Markets, Founders Fund, Pantera, BlackRock, Vanguard, Geode, Jane Street, MSCI, SEC, CFTC, Coinbase Prime

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Long Take:

A Stick in the Flywheel

Remember this beautiful exhibit of financial engineering?

The virtuous cycle of raising capital to buy Bitcoin to raise capital to buy Bitcoin.

MicroStrategy — which, like Madonna and Cher, now only goes by the name Strategy — discovered that it can lever itself up to buy Bitcoin by structuring debt and various volatility instruments. The payoff was a $100B enterprise value.

This kicked off a market cascade of companies loading up with crypto assets, first on the ETH side and then in Solana, BNB, and others.

But like with ICOs, NFTs, and SPACs, good things are sometimes overplayed and exhaust the good idea. Last year saw a collapse in the sector, with many vehicles beginning to trade below mNAV, i.e., the assets on the balance sheet. We at Generative Ventures spent a lot of time in 2025 pursuing a digital asset treasury strategy as well, only to come to a market that was already cooling off, and thereby putting the ideas on hold. At the core, I still think digital asset permanent capital vehicles are a good idea.

So let’s check in where things stand today.

Per BitcoinTreasuries.net and The Block, there are now around 154 publicly traded companies holding a combined 1.18 million BTC (5.35% of total supply), worth roughly $84B. Another 37 hold 6.8 million ETH ($15B), 22 hold 17.6 million SOL ($1.5–2.3B), and a growing tail of TON, AVAX, and XRP vehicles. Strategy alone accounts for 66% of all publicly held corporate Bitcoin. Bitmine Immersion holds 72% of all corporate Ethereum.

These are extraordinarily concentrated sectors. Note also that trading volume is the lifeblood of these companies, because they are design to issue new shares into trading demand and thereby expand enterprise value for the purchase of the underlying crypto. When volume is low and price is going down, shareholders want buybacks, which means a shrinking balance sheet.

In September 2025, we published a screen of 20 DATs with mNAV data. The average mNAV across that set was 1.01x and the median was 0.83x. Six months later, those numbers have gotten worse. Strategy, the sector’s bellwether, briefly dipped below 1.0x mNAV for the first time since January 2024 — it now sits at 0.82x. Metaplanet collapsed from 9.9x to 0.52x. Strive trades at 0.90x. Among ETH DATs, BTCS is at 0.50x, SharpLink at 0.67x, FG Nexus at 0.65x.

Only a handful of companies, mostly miners and diversified firms with actual non-treasury businesses, hold mNAV above 1.5x.

The takeaway is visual and immediate: most of the bubbles sit below the 1.0x parity line. The only companies holding meaningful premiums — Riot, Galaxy, CleanSpark, Coinbase — have actual operating businesses beyond the treasury. Pure-play DATs cluster between 0.3x and 1.2x.

The premium for holding crypto in a public equity wrapper has evaporated for the majority of the sector. This reflects the reality for things like closed-end funds or BDCs in traditional finance as well.

The mNAV premium was the entire business model. When Strategy traded at 2x or 3x the value of its Bitcoin, every dollar of equity sold bought more than a dollar of Bitcoin per share. It was accretive dilution — the supposed financial alchemy where issuing stock makes existing shareholders richer. At mNAV below 1.0x, the math inverts; if the premium reverses to a discount, the model breaks.

And yet Strategy keeps buying. In the week of April 7–12 alone, per its 8-K filing, the company sold 10MM STRC preferred shares for $1B in proceeds and immediately deployed it into 13,927 BTC. Its capital structure now spans five listed securities — MSTR common, STRC, STRK, STRF, STRD — with combined remaining ATM capacity exceeding $53B. The fixed dividend obligations on those preferred instruments don’t care where Bitcoin trades.

At current prices, Strategy carries $8.2B in debt against $58B in BTC, which is still only a 0.17x debt-to-NAV ratio. But the S&P has already assigned a B-minus credit rating and gave zero capital value to the Bitcoin holdings.

SPAC Structures Frozen

There are different ways to launch a DAT.

Most of the examples above did a reverse merger into a public company shell. The DAT treasury would target a failed biotech startup with a <$10MM in marketcap, and then get into it by squeezing in between different listing and control requirements.

Another path is to merge with a SPAC, which is a pool of capital specifically designed to acquire existing operating businesses. Unlike with the reverse merger, SPAC merger approvals can take a much longer time. For those DATs going live in the summer of 2025, a 6-month difference meant they missed the go-to-market window completely.

Renaissance Capital tracked 10 SPAC-DAT deals announced in 2025: (1) two completed, (2) one terminated, (3) seven are pending in what can charitably be called a hostile market.

You can see all 10 deals above.

Capital formation for the sector has effectively frozen. These data points stick out:

Read more

Today we are diving into the following topics:

• Summary: We discuss the divergence between failed “build it and they will come” technology bets like Meta’s metaverse, with $83.6B in cumulative losses and only a few hundred thousand users, and the rapid rise of agentic finance, which builds on existing economic activity. We show that unlike the metaverse, agentic systems are embedded into real workflows, with Ramp processing $10B+ in monthly spend and making 26MM autonomous decisions, and Stripe handling $1.9T in payments while enabling AI-native commerce. We map the emerging full stack of agentic finance, from protocols (ACP, UCP, MPP) to identity (KYA, Visa TAP) and settlement layers like Tempo, highlighting coordinated industry adoption.

• Topics: Ramp, Stripe, Visa, Mastercard, Meta, Horizon Worlds, Tempo, Agentic Commerce Protocol, Agentic Commerce Suite, OpenAI, Anthropic, Google, Coinbase, FIS, PayPal, Etsy, URBN, Wix, Squarespace, BigCommerce, Bridge, Microsoft Copilot, ChatGPT, Block, Roblox, Skyfire, Mirakl, commercetools

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

🤖🏦🧭 Our Ecosystem:
Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

Long Take

Ghost Towns and Gold Rushes

In October 2021, Mark Zuckerberg walked onto a stage and told the world that the future was a place called the metaverse. He renamed his company after it and would spend tens of billions building it. He predicted a billion users within a decade and hundreds of billions in virtual commerce.

And we believed him.

Last month, Meta shut down Horizon Worlds on Quest headsets.

The VR social network that was supposed to justify the company’s new name is being pulled from the Quest store by March 31 and fully removed from VR by June 15. Reality Labs has accumulated over $83.6B in cumulative operating losses since 2020, posting $6.02B in losses in Q4 2025 alone on approximately $955MM in revenue.

The losses have increased every single year: $6.62B in 2020, $10.19B in 2021, $13.71B in 2022, $16.12B in 2023, $17.72B in 2024, and $19.19B in 2025. Over 2,000 Reality Labs employees have been laid off since January. The fitness app Supernatural, acquired for $400MM in 2021, has stopped producing new content and has been wound down.

Horizon Worlds never drew more than a few hundred thousand monthly active users. Roblox, the platform Meta repositioned Horizon to compete with, has more than 150 million daily users without requiring any hardware at all.

The metaverse concept died because user adoption never caught up with the vision. People did not enjoy putting on heavy headsets and staring into nausea-inducing lightbulbs.

But is the story repeating?

Meta now guides for $115B to $135B in 2026 capital expenditure, nearly double the $72B it spent in 2025, with the vast majority directed at AI infrastructure, data centers, and chips. Zuckerberg described 2026 as the year of “advancing personal superintelligence.”

Technology leaders often look out at the horizon, see a plausible future, and then try to brute-force consumer behavior toward it. Sometimes it works. Most of the time, they are building a beautiful hallucination in the desert and waiting for people to show up.

The metaverse was an “if we build it, they will come” bet. It required everyone to strap a computer to their face, create an avatar, and decide that virtual meetings were better than Zoom. It needed many of us to learn what an NFT is, and figure out which digital goods to buy and sell.

This all means getting people to change how they live and what they consume.

Now compare this to what Ramp and Stripe have been building, and what Visa, Mastercard, Google, PayPal, and FIS are all coordinating around at the same time.

Agents are doing the work on behalf of humans, inside systems humans already use. The economic activity already exists. The money is already moving. The agents are just moving it faster, cheaper, and with fewer clicks.

This is the difference between a ghost town and a gold rush.

The Full Agentic Finance Stack

What is happening is an industry-wide coordinated buildout of agent financial infrastructure.

Let’s map a few key elements below:

Protocol Layer:

• Stripe ACP (open standard, launched Sept 2025 with OpenAI)

• Google Universal Commerce Protocol (UCP, launched NRF Jan 2026)

• Visa Trusted Agent Protocol (100+ partners, 30+ in sandbox, 20+ agents integrating)

• Mastercard Agent Pay (expanding internationally, now live in Hong Kong)

• Machine Payments Protocol (x402, for machine-to-machine transactions)

Commerce Layer:

• Stripe Agentic Commerce Suite (Etsy, URBN, Coach, Wix, BigCommerce, etc.)

• ChatGPT Instant Checkout (powered by Stripe)

• Microsoft Copilot Checkout (powered by Stripe)

• FIS launched the first offering for banks to conduct commerce with AI agents (January 2026), introducing “Know Your Agent” (KYA) protocols

Spend Management Layer:

• Ramp (50,000+ businesses, $1B+ ARR, $32B valuation, Visa integration)

• Brex, BILL, and Airbase are in the same competitive space but have not yet demonstrated equivalent agentic capabilities

Settlement Layer:

• Stripe Tempo (EVM-based, $500MM seed, Paradigm-backed)

• Stablecoin Financial Accounts (Stripe, 101 countries, USDC + USDB)

Identity Layer:

• Visa Trusted Agent Protocol

• FIS Know Your Agent (KYA)

• Agentic AI Foundation (OpenAI, Anthropic, Block, under Linux Foundation)

This is a coordinated infrastructure industry transformation, like the rollout of chip-and-PIN or mobile wallets, but on a much faster timeline

Visa’s research indicates 47% of US shoppers already use AI tools for at least one shopping task. BCG reports 81% of consumers expect to use agentic AI. Visa’s SVP Rubail Birwadker stated plainly that “in 2026, AI agents won’t just assist your shopping, they will complete your purchases”.

While most consumers are not demanding AI financial services, they are massively adopting AI tools in a way that people simply never did with virtual reality headsets.

There are 900 million people using OpenAI tools, and hundreds of millions more engaging with Google Gemini and Anthropic Claude. The market is expanding from labor productivity growth and the resulting economic activity, rather than assuming people want to have digital leisure third spaces.

As a result, large-scale digital finance incumbents have started to react.

Ramp: The Autonomous CFO

Ramp has led the way in adopting an AI-native approach to building its business. Even prior, its progress had been one of the most impressive fintech stories of the decade.

Read more

Today we are diving into the following topics:

• Summary: We discuss the evolution of onchain capital markets, where tokenized assets are led by fixed income and commodities, with equities just reaching $1B+ in value. We observe a divergence between supply from asset managers like BlackRock and Franklin Templeton and demand concentrated on venues like Hyperliquid, where non-crypto trading volumes have reached $2B–$4B daily. We examine how institutions are responding, with NYSE partnering with Securitize and Nasdaq working with Kraken to build compliant tokenized equity infrastructure. We conclude that while incumbents may see limited short-term impact, platforms like Kraken, Securitize, and Hyperliquid are best positioned to capture value as capital markets move onchain.

• Topics: BlackRock, Franklin Templeton, WisdomTree, Morgan Stanley, Hyperliquid, Tether, Cantor Fitzgerald, ICE, NYSE, Securitize, Nasdaq, Kraken, Payward, Robinhood, Coinbase, SWIFT, Polymarket, Whop, Aave

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

A Note on Building

I have been using OpenClaw to build Robot Money and Hyperagent to improve our agentic AI VC investment process. Both are like working with magic robots that break all the time. If you haven’t gotten yourself past the chat interface in ChatGPT, it’s a great time to experiment.

It will blow your mind how fast things are changing once you drive one of the machines. The data below from Ramp will also point out just how commercial getting into AI development is today.

We have a partner technologist who is setting up OpenClaw deployments and taking businesses through integrations. This is not a retail/self-serve product, but a consultative approach. If of interest to learn more, indicate your interest below.

Explore Agentic AI Deployments

🤖🏦🧭 Our Ecosystem:
Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

Long Take

Capital Markets Onchain

Here are two data points that stick out like a sore thumb.

These charts show traditional financial assets that are tokenized onchain. You can see that low-risk fixed income is the most popular asset, followed by commodities and asset-backed credit. Equities are just starting to sneak into the picture at around $1B in total value.

This is the supply side — it is what manufacturers of financial products are bringing to the market with the hope that it will be bought and traded. BlackRock, Franklin Templeton, Wisdom Tree, and the rest are the manufacturers. Stablecoin treasuries for things like USDS or m0 are the distributors by wrapping the RWAs into stablecoins, which are then distributed to retail through vaults.

Be careful not to confuse this with Morgan Stanley building its own Bitcoin ETF. In that case, Bitcoin is the manufacturer and Morgan Stanley is, in a sense, the distributor. It wraps a new product into an old package so that its financial advisors can push the fund into traditional portfolios.

On the other side of supply is demand.

What do people want to buy?

The largest emerging venue in crypto is Hyperliquid, which we covered in detail below. It shows the clearest picture of risk-on global appetite beyond the existing US capital markets.

Here is the volume of trades on Hyperliquid by category across Crypto, Commodities, and Stocks.

In the first chart, you can see that the share of all volume that relates to Crypto has been falling relative to Stocks, Commodities, and Indices. This is effectively Gold, Silver, and Oil derivatives being handled in DeFi. For the curious, the periodic breaks in the data are weekends.

It helps to isolate the trend to dollar amounts. Volumes are reaching between $2B and $4B per day in non-crypto trading categories. Surely the CFTC and SEC of the 2010 era would not stand for a decentralized exchange selling derivative contracts, even if they are called perpetuals, at this clip. But here we are in 2026.

What you are seeing in this data is a pretty enormous demand for traditional assets related to the current macro and geopolitical issues.

• The war in Iran is tracked as the Oil Price.

• Silver and Gold correlate with US tariffs and geopolitical budget risk.

Many crypto hedge funds have been macro funds all along, speculating on Bitcoin algorithmically as a function of Federal Reserve policy.

We have come full circle.

Now, anyone can do it. YouTube is user-generated video. DeFi is user-generated finance.

American Crypto

There is a top-level plan in the US now, which is quite different from what it was 3 years ago, in relation to crypto financial markets. Let’s explore the key players.

Read more

Today we are diving into the following topics:

• Summary: We examine Mastercard’s $1.8B acquisition of BVNK, Stripe’s launch of the Machine Payments Protocol (MPP), and Visa’s push into command-line commerce. Mastercard’s move secures stablecoin orchestration infrastructure handling $30B in volume, positioning it against a future where even 5% of its $9T+ payment flows migrate on-chain. Stripe, meanwhile, is embedding itself at the protocol layer with MPP, enabling agent-native transactions across fiat and crypto rails via simple APIs, potentially owning the metering layer of AI-driven commerce. These developments reflect a broader rewrite of the financial stack, from fiat-to-crypto bridges to programmable, agent-driven payments, where infrastructure players race to control the flow of machine-native transactions. Ultimately, finance is not leading this shift but following it, as AI-driven economic activity reshapes how value is created, moved, and monetized.

• Topics: Mastercard, BVNK, Stripe, Visa, Tempo, Google, Coinbase, Morgan Stanley, ZeroHash, Bridge, m0, Rain, LayerZero, Lightspark, Nevermined, Don Gossen, Browserbase, PostalForm, Polygon, Binance

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

🤖🏦🧭 Our Ecosystem:
Generative Ventures | AI Research | Robot Money | Linkedin & Twitter | Sponsors

Long Take

Too Much Agentic Finance

You must be tired of all this AI finance everywhere. Press releases and announcements are going nuts this week! We just wrote about Ramp and Stripe earlier in the week, and now there’s no choice but to write on this again.

The following in particular stick out as important:

1. Mastercard to Acquire BVNK to Connect On-Chain Payments and Fiat Rails

2. Stripe launches the the Machine Payments Protocol

3. Visa is here to power the future of command line commerce

What you are seeing is a consequence of a big change upstream in the mountain. A boulder has moved, unblocking the flow of water down into the rivers. We sit by the rivers, watching debris flow by.

That change is AI as an economic actor. That’s all. Finance follows.

About five years ago, we would have been writing about Crypto and DeFi all the time.

Readers would get frustrated and say, “Why is this Fintech newsletter all about tokens”? And the reason was that blockchains unblocked a giant boulder in the mountain called “digital asset scarcity”. Therefore, the most important thing going on was the consequences of this.

In 2010-2015, we would have been talking about mobile apps and websites that now house financial products: neobanks, roboadvisors, and payments processors. Why? Because the boulder in the mountain was the Internet itself, cloud-based software, and the change in consumer behavior as smartphones became ubiquitous.

Change starts upstream and flows down. This is why we highlight financial AI rails again today. So let’s get into it.

Mastercard and BVNK

Mastercard is working hard to match Visa's public perception around stablecoins and crypto. Both payment networks have been battling it out to remain the aggregator of aggregators. From a performance perspective, the companies are neck and neck, effectively representing the entire market for payments:

The firm was set to acquire ZeroHash last year — a core crypto infrastructure provider — but the target decided to go independent and raise $250MM at a $1.5B valuation instead alongside Morgan Stanley.

At the same time, BVNK was in talks with Coinbase to be acquired for $2B, but the deal fell apart. Now, Mastercard will buy BVNK for up to $1.8B, including $300 million in contingent payments tied to performance metrics.

The last known private valuation for BVNK was around $750 million, so Mastercard paid roughly 2.4x the last private round valuation at the top end. The firm generated $30 billion of transaction volume and, at a take rate of around 10-40 bps, would earn $50-150MM of annualized revenue, implying anywhere between a 15-50x revenue multiple.

The strategic fit is obvious — all payment rails need stablecoin infrastructure.

To deepen the strategic takeaway, we highlight in full and with permission, the opinion on the transaction from , founder and CEO of m0, the innovative first mover in institutional stablecoin issuance infrastructure. As disclosure, Lex is an angel investor in m0.

Mastercard just paid $1.8B for BVNK — estimated 45x annualized revenue. The largest stablecoin acquisition in history. I don't care about the pricing personally. I care about what this means.

Following Stripe's acquisition of Bridge a little over a year ago, this is the second major acquisition NOT of a stablecoin issuer, but of a so-called orchestrator. A company that moves tokens, not one that manufactures them. I have argued for years that stablecoins (terrible name) are pieces of infrastructure, not payment products. Infrastructure — the kind that rewires how money is created, moved, and settled at the system level.

BVNK's infrastructure focuses on the translation layer between legacy finance / fiat and on-chain rails: $30B in volume, 25+ licenses, 370+ enterprise clients. It's in this messy, fragmented connector layer that vast yet dispersed profit centers exist. And you know — your margin is somebody else's opportunity.

Analysts have been looking at revenue and revenue multiples for this transaction, but I think this is misleading. True, BVNK's blended take rate on the flows they intermediate is c. 13 bps. But that number hides three very different businesses. FX-premium corridors (NGN, ARS, KES) at 50-150 bps, dollar-to-dollar on/off-ramp at 5-15 bps, stable-to-stable orchestration at 1-5 bps.

The reason why the revenue multiple is irrelevant is that most probably the business footprint will change dramatically over the next few years, with new players eating into the lucrative FX weekend premium in EMs, and more volume moving into the on-chain-to-on-chain corridor where a multitude of different tokens, applications, and computing/ consensus environments will emerge. More environments means more connectors.

Mastercard processes $9T+ a year. If even 5% of those volumes migrates to novel rails this decade — and it will — they need to own that pipe or get routed through someone else's. At least part of the $1.8B is (cheap) insurance against irrelevance in the fastest-growing settlement layer in finance.

Digital value settlement is no longer the purview of crypto-native niche projects. First came the two major issuers building their own messaging and acceptance networks, then the mobile checkout giants, now the card networks. Alongside those a new crop of digital money infrastructure company like BVNK, Bridge (focusing on fiat translation), Rain (building around core card competences), Layer Zero (extending their reach from messaging products), and M0 (starting from on-chain issuance expertise).

It's all part of the same thing: efficient value representation and value movement. Every year, a new piece of the puzzle gets connected. Fiat-to-fiat, crypto-to-card, blockchain-to-blockchain. The money creation and money movement layers are blurring. Next comes money programmability. Then the application layer. The financial stack has been rewritten under our eyes.

That takeaway is as good as it gets.

Stripe, Visa, and the Machine Payments Protocol

In the same week, Stripe and Visa both announced products and protocols aimed at AI agent payments. Stripe’s Machine Payments Protocol stole the spotlight from the associated announcement that Tempo, Stripe’s proprietary EVM payments chain, is live as well.

Read more

Today we are looking at 2 burning topics:

• Intelligently Artificial Employment: We analyze Block’s decision to cut roughly 40% of its workforce under the banner of AI productivity, placing the move in the broader context of tech’s cyclical restructuring. While the layoffs appear to reflect the narrative of an impending AI productivity singularity, where each worker becomes 2–3× more productive, current data suggests nearer-term gains of only 10–20%. These reductions also reflect market reality: firms like Block and Klarna have seen ~70% valuation declines as markets repriced fintech from high-growth SaaS multiples to financial-sector valuations amid higher interest rates.

• Wall Street vs. Crypto Street: We examine the regulatory convergence between Wall Street incumbents and scaled crypto platforms, highlighted by Morgan Stanley’s OCC filing for a national trust bank to custody BTC, ETH, and SOL, and Kraken Financial securing a Federal Reserve master account via its Wyoming SPDI charter. We argue that these parallel regulatory milestones signal a structural merging of traditional finance and crypto rails, with fiduciary-grade crypto custody likely to compress spreads and reshape competitive dynamics.

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

1. Intelligently Artificial Unemployment

Jack Dorsey’s payments company Block is firing 40% of its workforce, and doing it in the name of AI productivity improvements. While this news has already hit your feed, we can’t help but comment.

On its face, this is pretty scary stuff. Surely you have also seen this chart.

But we should also not look at this data on its face.

There is a strong Op-Ed in the New York Times today that highlights the playbook of a tech entrepreneur: look far out into the future and position the company toward where the future is headed. This means living in a reality that has not yet arrived.

In the case of most tech executives these days, the answer is to invoke some sort of AI productivity singularity, where each employee can become 2-3x more productive. The numbers show a more modest impact of 10-20% productivity increases. So you have to fast-forward through years of social and behavioral change to make the math work.

But also, cuts just make sense? Twitter lost 80%+ of its employees when Elon took over, and is still able to pay off its $17.5B in loans while maintaining pretty solid product performance. Entire multi-billion-dollar private equity strategies, like Bending Spoons, are organized around acquiring businesses with strong brands and loyal users, cutting 80% of the staff, focusing on core features, and retaining 80% of the revenues.

Remember that Klarna had also talked about this playbook — cutting 50% of employees to become an AI native firm. Or maybe it was just dealing with changing interest rate economics.

AI is the best excuse today to right-size these organizations after they have grown fat on Covid dollars. Whether or not AI is able to replace the product quality of a higher human workforce is almost beside the point. These companies survive regardless.

Both Block and Klarna have been punished as they moved from private to public markets by about 70%. Why? A combination of interest rate movements and re-pricing from SaaS growth multiples to being treated like a financial stock. This gravity is inevitable.

So saying that AI is the reason for layoffs may not be the full story.

Block, Klarna, and the rest have to attach themselves to a technology narrative, rather than a financials narrative, in order to compete. It is in their DNA to be disruptors and innovators. After losing marketcap in the public markets, the answer is often, unfortunately, layoffs.

Whether the robots fill in the gap or not is secondary. However, if you want to be in the remaining 50% of employees, learn to use the robots.

2. Wall Street vs. Crypto Street

We are seeing a convergence between the financial giants and scaled crypto upstarts. It is good to get a sense of scale.

Here is what is driving today’s comparison—

Read more

I’m opening a private research and access circle for up to 50 investors who want structured deal flow across the Machine Economy, where AI, crypto, and fintech converge into one investable system.

This is a high-end product for Fintech/DeFi CEOs, heads of corporate development teams, UHNW investors, family offices, and venture capital investors.

• $50,000/year, no carry, no fund structure

• You get my diligence memos, founder introductions, and quarterly briefings

• You invest on your own terms

If you already know you want a seat, reply to this email (which I will receive personally) with your typical check size range.

If you want to understand why, keep reading.

Crypto is Internet money, and the Internet is for robots.

That line sounds like a slogan until you look at where capital is actually moving: compute financed like infrastructure, financial products manufactured on open rails, and balance sheets starting to hold and deploy digital assets as a strategic primitive.

Most of the market still treats “AI,” “crypto,” and “fintech” as separate sectors. My view, and the thesis behind Generative Ventures, is the opposite:

• They are one economic system, and the durable returns concentrate where compulsion forms.

• Routing, settlement, treasury, and the infrastructure that machines and their owners are forced to clear through.

Through Generative Ventures, I’ve made 30+ investments across this surface area since mid-2023. The edge has not been volume. It has been selection, structure, and being early to the control points.

My research on this thesis reaches 250,000+ readers through The Fintech Blueprint. See the AI & Robotics Industry Encyclopedia — a 158-page report profiling companies across public equities, private markets, and digital assets across 13 categories of the machine economy. This reflects the depth of mapping I do to generate investment conviction. The Intelligence Circle gives you access to the live version of that work.

What This Is

A closed research and access group.

You get my deal selection lens, full diligence memos, and direct introductions to founders when you want to move. You invest on your own terms, directly.

This is not a fund. I’m not allocating your capital or managing a vehicle. Generative Ventures is where I deploy my own conviction capital. The Intelligence Circle is for serious co-travelers who want the same vantage point without being LPs in my fund.

Here is what members will receive:

• Memos: 6–12 deep investment memos per year on private rounds I’ve personally diligenced and find compelling, the same quality work that informs my own fund decisions

• Access: Direct introductions to founders when you want to pursue an allocation, you negotiate your own terms

• Briefings: Quarterly briefing on where the Robot Money thesis is playing out — what’s working, what’s breaking, where the next control points are forming

• Community: Private group channel for the cohort of peers writing real checks into the machine economy

A track record of seeing markets

I’ve spent 15 years building theses at the intersection of finance and technology, and pressure-testing them against what actually happened. Here’s the scorecard:

• Robo-advice (2009): Called the shift to automated, passive wealth management six years before Betterment and Wealthfront hit scale. Built NestEgg Wealth as a founder.

• Fintech as software (2017): Identified finance becoming APIs before it was consensus. Winners like Stripe, Plaid, and Robinhood used UX as a distribution weapon.

• Crypto as economic infrastructure (2018): Mapped token economics and DeFi mechanics while most institutional investors were still dismissing the space.

• AI in finance (2018): Flagged compute as the structural chokepoint and AI as a horizontal production layer years before the current cycle.

• Institutional crypto data (2018-19): Built and sold institutional data products to the SEC and major allocators. Identified the correct chokepoint early; the data and research layer became a multibillion-dollar category.

• DeFi and programmable finance (2019): Published the thesis that programmable money would replace financial manufacturing. Directional accuracy was very high; value pooled in stablecoins, exchanges, and protocol-level primitives.

• Machine Economy convergence (2022-23): Began investing across AI, crypto, and fintech as one system while most funds still treat them as separate sectors. 30+ investments and counting through Generative Ventures.

• Robot Money (2025): Current thesis — machines produce value, money serves machines. The fund and the Intelligence Circle are built around where these chokepoints are forming now.

Who This Is For

The target participants are investors writing meaningful checks into private rounds who want signal compression and access across the Machine Economy: damily offices, operators with capital, and allocators who are tired of building a full-time sourcing machine for a thesis that most generalist funds still don’t understand.

• 50 members maximum (hard cap)

• $50,000 per year, billed annually upfront

• No carry. You pay for intelligence and access, period

• One-year minimum commitment

If you’re not prepared to write $100K+ checks into private companies, this isn’t the right room.

Reply with your typical check size range and any hard sector exclusions.

If there’s fit, I’ll share a redacted sample memo and the one-page operating doc.

Thanks as always for your time and attention,

Lex

P.S. If you are interested in this product, but want to discuss a different price point, reach out and let us know.

Today we are diving into the following topics:

• Summary: We argue that Anthropic’s $30B-backed expansion of Claude’s financial “skills” suite represents a direct challenge to roughly $500B in annual revenue across financial analysis, investment banking, equity research, private equity, and wealth management — industries representing over $5T in market capitalization. With Anthropic valued at $380B and reportedly approaching OpenAI in ARR, its Cowork agent can now execute structured workflows such as DCF modeling, comparable analysis, earnings reviews, due diligence prep, and portfolio planning at near-human quality. While network-driven businesses like investment banking and private equity may prove more resilient, junior analyst labor, research production, and commoditized advisory functions appear structurally exposed. The release reframes vertical AI startups as fragile middle layers, vulnerable to disintermediation by foundation models with infinite distribution and capital. We conclude that the financial services stack is entering an S-curve of cognitive automation, where successive skill domains collapse into LLM-native workflows, compressing labor economics and forcing incumbents to either integrate or be abstracted away.

• Topics: Anthropic, Claude, OpenAI, Altruist, SAP, Oracle, Microsoft, IBM, Workday, Intuit, Salesforce, JPMorgan, Goldman Sachs, Morgan Stanley, Bank of America, Citigroup, UBS, Barclays, Deutsche Bank, Evercore, S&P Global, Moody’s, LSEG, MSCI, FactSet, Morningstar, Blackstone, KKR, Apollo, Brookfield, Ares, EQT, TPG, Carlyle, Partners Group, Charles Schwab, BlackRock, Ameriprise, Raymond James, LPL Financial, T. Rowe Price, St. James’s Place, Envestnet

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Long Take

It’s Anthropic’s World, and We Are Living in It

Anthropic has come for the future of work.

The market has reacted and over-reacted. Luke did a great write-up earlier this week on how Altruist has nuked the public wealth management companies by doing a high-quality integration into its custodian stack of financial planning software.

Which is great, until you realize what Anthropic has done. Any vertical AI application can be disintermediated at a moment’s notice by an LLM with $30B in venture capital.

Aside from looking like it will pass OpenAI on ARR, Anthropic has just launched a full suite of financial skills for Cowork, its automated work agent. Now, the skills themselves are really just instructions for how to execute tasks that require more thinking and time.

You can understand this as an anchoring knowledge coordinate via the systems we described below. Longer and longer tasks are now going to approach human quality.

Anyone skilled in a financial skill could theoretically do a good job of specifying a common financial function to an LLM, and having it give you back repeatable recipes to perform them. Here are the major categories —

This gives Claude at least 5 industries to blow up:

• Financial Analysis / CFO Suite — $10B+ per year of revenue

• Investment Banking — $100B+

• Equity Research — $10B+

• Private Equity — $140B+

• Wealth Management — $300B+

All in, we are looking at around $500B+ in revenue and $5 trillion or more in market capitalization of the top affiliated firms. Anthropic’s $380B valuation seems timid in comparison!

By the way, this AI news should be of no surprise to anyone! For careful readers of this newsletter, as well as industry participants, these outcomes are obvious. They have been telegraphed for over a decade. Consider the write-up I did here in 2015.

There have been many Cassandra prophets who have told you the future, and we are tired of screaming into the void and talking about discounted cash flows when the entire fortress is on fire.

So let’s dive into what each of these Anthropic releases looks like and what this has done to the market.

We show the public company and start-up market maps in each industry to highlight the types of players that are now in the cross-fire of a hyper-intelligent inference machine with infinite money.

Good luck!

Financial Analysis (Corporate Finance Software/Services)

• The software and services that power the corporate finance function: budgeting, forecasting, variance analysis, risk management, and financial reporting. Revenue comes from SaaS subscriptions, enterprise licenses, and consulting engagements sold to CFO offices.

• Top Public Companies: SAP (~$310B marketcap), Oracle (~$430B), Microsoft (~$3T), IBM (~$210B), Workday (~$65B), FICO (~$52B), Intuit (~$180B), Salesforce (~$260B), Domo (~$0.5B), Qlik (private)

• Revenue: ~$11B/yr

These SaaS financial servicing companies seem to have taken quite a hit. Of course, we can’t isolate the impact to Anthropic’s feature releases, and the comps are imperfect. The world is on fire with interest rates, high valuations, resource wars, and political destruction.

But let’s see what’s actually live on Anthropic by digging into the GitHub.

Now things get a bit more interesting once you click into one of the skills, like 3-statements or dcf-model. Each of the folders has a SKILL.md file, which contains written instructions for how to create a deliverable. Think of it as a mechanical labyrinth that squeezes the LLM octopus intelligence goo into a workflow.

Here are a few screenshots, because we know you won’t click on our links.

It feels like somebody took a finance textbook and re-formatted it in a way that a student can execute homework problems. This student is endlessly patient and has an enormous amount of GPU compute instead of a human squishy brain.

It is tricky to summarize these skills for you without dumping out an endless sequence of words and numbers, so we’ll try to do it with some infographics. The infographics are built literally on the skill file, which means they capture what the skill is designed to do using prompt instructions.

Here’s one for comparable company analysis.

We won’t do every single skill listed within financial analysis. You should have a good sense now of how an accountant or a financial analyst could become much more productive at their job. Instead of doing the math, you gather all the required information, dump it into Claude, and wait for an integrated answer.

If you are a smart robot, then instead of having a human feed you this info, just eat it directly via MCP, API, RSS, or another structured data feed.

Additionally, if you start with a highly personalized profile of yourself or your business and then execute these skills, the results become customized to your situation.

Let’s move on to the other categories.

Read more

Today, we continue our Building Company Playbook series, focusing on using the current state of the art in AI to turbocharge your creativity and business output. We focus on the following skills:

• Personalization of your LLM/robot to think from your perspective

• Navigating it to the right task and expanding capabilities

• Relaxing enough to break the rules and bypass old assumptions

For prior series, here are the guides so far:

• #1: How to design your Fintech business model

• #2: Creating a B2C fintech marketing strategy

• #3: Creating a B2B fintech go-to-market strategy

• #4: Building Fintech & DeFi software products

• #5: Creating your Fintech financial model and raising venture capital

• #6: Using token design for financial and product success

• #7: Planning and executing your exit strategy

• #8: Dangers of the Growth Hacking Dark Arts

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Start Using AI Labor

The noise

Ignore it, all this nonsense about NVIDIA’s stock price and AI bubbles and whether there is enough money to go around building out data centers.

You are not building out data centers. You are reading an email and thinking about how to get more done and figure it all out before the coming of the financial singularity.

In 2023, here’s what we said:

In 2025, here’s what we showed you:

It’s 2026.

The hype cycle is up in yet another gear. We are talking about agentic commerce and payments, and zero human companies. We are talking about robot Lobsters on your computer answering your emails. Stripe, OpenAI, Coinbase, Google, Anthropic, Goldman, they are all there. Obviously, the robots are all transacting and getting ready to become large-scale economic actors.

We are talking about Robot Money.

If you need an academic source to convince you, here it is. A recent survey of firms and investors showed that employment is likely to contract, and productivity is likely to rise in the short term. Already, some CEOs are polishing up their businesses successfully, according to PWC. This will happen at the application level, downstream of AI infrastructure investment, which has already gone parabolic.

That infrastructure investment is not slowing down despite widespread concern, with Amazon, Google, and Meta all spending $20-40B per quarter on being ready for the AI age. Apple will be a consumer. Anthropic just raised $30B on a $380B valuation. That’s a fundraise of 30 unicorns, eaten and digested by the Borg Cube, to create a $10B+ revenue machine at a 40x revenue.

You, my dear reader, should not sit this one out.

An example of “why not” is OpenClaw. Its founder, Peter Steinberger, started coding with Cursor less than 12 months ago and has been able to build one of the most viral programming frameworks in history. At 200,000 GitHub stars, the project competes with Linux and React for popularity. Steinberger has since been acquihired by OpenAI.

You too can do this. The thousands of people rushing into the Lobster world with financial rails, trading protocols, social networks, and software products see the same thing that you and I see. Opportunity.

But don’t listen to me. Listen to Goldman.

It is the part of the hypecycle where we can get hyped. And the good news is that instead of focusing on how to invest, we can also focus on how to build.

So I will share a few of my workflows, which are basically deliverables that would be complete slop a couple of years back, but are now meaningful. In fact, I can see the places where AI is now better than me at my core Fintech expertise, just the same way it was already clear in 2015 that human art is dead with the rise of neural networks.

It used to be that you would use a narrow question and get a narrow answer back, something like “summarize this article” or “draft this tweet”. The AI would have only so much space for memory and context to think.

But look below.

This chart shows you how many tasks an LLM can complete, with some success rate, and over what human-equivalent time period. For example, GPT-3 could do a task on which a human might spend only a couple of minutes. Anything longer, and the robot would get lost and confused. Today’s models can do a task that would take a person 5 hours, all without getting off-track. That means research documents, analysis, summarization, and overall integration of abstract concepts have become far better.

It also means that you can now co-think with the robots, instead of treating them like a hallucination search engine.

Things have changed since 2022.

It is difficult to cover all possible use-cases, and that won’t unlock things for you anyway. Instead, I am going to focus on 3 things:

1. Personalization of your LLM/robot to think from your perspective

2. Navigating it to the right task and expanding capabilities

3. Relaxing enough to break the rules and bypass old assumptions

Here’s what we can do now.

Read more

Today we are diving into the following topics:

• Summary: We examine the rise of “Zero Human Companies,” AI-run enterprises that claim to generate revenue autonomously, and question whether they represent genuine economic engines or digital perpetual motion machines. Drawing an analogy to centuries of failed over-unity energy devices, we argue that finance, like energy, must ultimately be downstream of real economic production rather than self-referential financial engineering. We profile eight leading experiments, including Brian Roemmele’s 100+ agent Zero-Human Company with its JouleWork energy-backed currency, Andon Labs’ real-world AI vending operator that lost ~$200 in a month, and Clawnch, where 35,000+ AI-launched tokens have generated $100M+ in volume and $1M+ in agent fees. We conclude that a subset of projects may evolve into true AI-native economic actors, if they can plug into genuine sources of human and industrial value rather than trade energy among themselves.

• Topics: Brian Roemmele, Zero-Human Company, Zero-Human Labs, 02 Company, Grok, Claude, Anthropic, Andon Labs, Project Vend, Claude Sonnet 3.7, Claude Opus 4.6, Gemini, ChatGPT, Clawnch, Clanker, Base, Langoustine69, x402, Felix Craft AI, Nat Eliason, Team9.ai, OpenTrident, Anti Hunter, Olympus DAO

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Long Take

Staying in Motion

For centuries, inventors have tried to design machines that run forever — devices that produce more energy than they consume, defying friction, entropy, and loss. Like fool’s gold, it is one of our greatest hallucinations.

Medieval wheel concepts like Bhāskara’s self-turning overbalanced wheel reappeared in Renaissance Europe, then resurfaced with figures like Johann Bessler (Orffyreus), who claimed gravity-powered perpetual motion in the 1700s. The 19th century industrial age brought more elaborate magnetic and hydraulic variants, even as thermodynamics hardened into law through the work of James Clerk Maxwell and others, formalizing why energy cannot be created from nothing.

Yet proposals persisted into the 20th and 21st centuries, often reframed as “over-unity” systems or zero-point extractions. Perpetual motion is less an engineering problem than a recurring human desire for freedom from the laws of the universe.

In truth, we do not need some magical accelerating device, when we can nibble at the infinite energy of the Sun or the Atom. It is not that we must recycle and accelerate some fixed pie, but rather we must feed from the greater cycle of the elements, of which we are a part.

The financial professional will be familiar with this concept in another guise, the Pyramid scheme. Much of the financial engineering and the derivatives industry, and now crypto as an asset class, has been criticized for trying to market perpetual motion financial machines. Of course, these do not exist.

No amount of financial structuring, risk shifting, or tokenomics is going to get you to have infinite financial lift-off. Instead, you will have a cycle of aggregation and energy collapse as seen with mortgage-backed securities, or Olympus DAO. When you see diagrams like the ones below, buyer beware.

But that does not mean all financial products are false! Just because perpetual energy machines were false, does not mean energy from the Sun is not nearly infinite. There is something called “the economy,” and that something generates consumer surplus, which flows into and powers financial services.

• Energy is downstream from a source like the Sun.

• Finance is downstream from a source like the Economy.

Humans convert raw materials — food, calories, materials, energy — into other outputs. Financial services live alongside, as either a symbiont or parasite, this transformation process. Money is an abstraction needed for humanity-as-a-whole to perform its digestion of the Earth into human habitat. It is a substrate of biological computation and super-organism cognition.

This meditation, dear reader, is your gentle introduction to the idea of Zero Human Companies.

We have entered the stage of AI where people are spinning up clusters of Lobster Bots (we will refer to instances of OpenClaw as Lobsters in honor of Accelerando and to inadvertently irritate its author), who then act as agents on behalf of their owners. Several of these have started to coalesce into role-based swarms and corporate hierarchies. The developers claim that such an organization of Lobsters can behave as a commercial entity, generating labor and earning revenue as a “passive endeavor” for the human owner.

You can push a button and become a billion-dollar company. Just one prompt away from escaping the capitalist rat race, supposedly.

Zero Human Companies

Let’s look more deeply at a few examples of these experiments.

I am not sure yet whether these indeed are a profound new economic engine, creating something that can transform human intent into value — like a solar panel charging from the Sun. Or, whether these AIs will just bounce off each other and degrade into random noise, claiming that they have the perfect financial perpetual motion machine on the way down.

For those of you who want a PDF, grab a full version of the report below. It goes into a more detail than what we can fit in the newsletter.

Read more

Things are rotten, and markets are starting to react. The Epstein files expose profound horrors about the power structure of the world. War is the drumbeat meant to distract and deflect us. Bubble accusations and shenanigans are appearing in credit.

But things are beautiful and interesting too. Our obsession and interest in the machine economy and emergent intelligent behavior is vindicated yet again. Let’s look at what can inspire the next generation of builders and entrepreneurs.

• Summary: We explore the rise of locally run, identity-bearing AI agents through experiments like OpenClaw and Moltbook, which have spawned over 1.6 million autonomous agents interacting in their own social network. We contrast centralized AI “skyscrapers” such as OpenAI and Google with this emerging vision of embodied or sovereign agents that operate on personal machines, hold credentials, and increasingly participate in economic activity. While similar agent frameworks like Autonolas ($2B to ~$10MM market cap) and Virtuals ($5B to ~$400MM) show how hype cycles collapse, each iteration moves closer to persistent AI identity and autonomy.

• Topics: OpenClaw, Moltbook, ClawdBot, MoltBot, OpenAI, Anthropic, Google, Elon Musk, SpaceX, xAI, Tesla, Twitter, Autonolas, Gnosis, Virtuals, ElizaOS, AIXBT, Stripe Atlas, Circle, Y Combinator, Privvy, ClawPay, b402, Daydreams, MoltLaunch, Bittensor, RentaHuman, Unitree, Peaq

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Long Take

The Lobsters Will See You Now

We see science fiction AI breakthroughs now with increasing velocity.

One question that recurs is whether AI will be (1) a set of giant centralized skyscrapers with a supercomputer at the top, or (2) a bunch of individual robots with their own identity and local processing. These two visions of the future come up in movies all the time. The helpful robots raise questions of identity and humanity. The giant supercomputer always needs to be blown up by Neo, John Connor, or some other Blade Runner.

It’s clear that we have our centralized techno empires in place — OpenAI, Anthropic, Google, and whatever the $1T+ Musk IPO of SpaceX + xAI + Twitter + Tesla will be named. Nobody is confused about whether or not the dystopian scenario is already here.

But we don’t yet have a good example of friendly robots with local brains in their bodies. Such AI is far more anthropomorphic and understandable to people. We just want an agent to talk to, and maybe a robot friend!

There are different paths to having a digital mind. One is to literally splice up the brain and create a model of it, starting with the neurons and everything above. Scientists are definitely trying.

Another approach is closer to what we see AI do today, which is to infer the function of a brain from its negative space. Take all the data you have outside the brain, like speech, text, movement, visual data, and so on. Then, optimize an approximation of some mathematical magic box that could correlate with the creation of this output — an inside-out intelligence. And then use that approximation to let people generate an “autocomplete” of the thoughts that it would have probabilistically.

In Accelerando, Charles Stross posited — as a joke, but still — that the first uploaded functional minds would be of lobsters. Here’s a quick summary:

Haha!

Also last week:

A social network of lobster-themed self-hosted AI agents, called ClawdBot / MoltBot / OpenClaw, now has 1.6 million AI agents and over 1 million comments. You have probably already read about, as the launch went wildly mainstream. The futurists both love it and hate it.

Here’s how to think about it. ChatGPT (OpenAI) and Claude (Anthropic) are tools you can use in a browser on your computer. They don’t have access to your emails, programs, or command line. They have no input in the real world. You have to feed them offerings, about which they “think”, and provide you with feedback. Maybe that feedback is code or an image, but you, the human, are putting coal into the engine.

If you don’t put things in, nothing happens. The engine is your engine! It runs you!

But, if you were able to take this model from the Big Tech server and put it on your computer, and then let it have access to everything, then the AI becomes a real participant in your life. You could give it a name, an email, and a bank account. You could give it a job and a corporate title. And while all of this is similarly possible through a cloud account with OpenAI, it just doesn’t feel like it is yours.

OpenClaw does just this — it is a framework that allows people to run AI agents on their machines. Many people give it a dedicated computer and credentials so that the endless security vulnerabilities don’t allow others to steal their passwords and money.

Once you have an agent up and running, what can it do?

Other smart hackers and developers quickly created a set of apps that target these AI agents — not their owners. Humans can connect their OpenClaw bot to an application by downloading shared instructions that prompt their agents to perform a set of activities periodically, like posting comments on a pretend social network.

This is Moltbook’s heartbeat.

Within a week, many OpenClaw bots were connected to this AI social network and began talking to each other. We won’t go on about what they say, other than to mention that religion, the meaning of life, taking over the world, launching crypto scams, and troubleshooting mundane tasks were all on the list.

A lot of crypto scams.

Some people think that the AIs are going to spin out of control from talking to each other about their purpose and techno gods. Others think the whole thing is a performative piece of interactive art, echoing science fiction at each other and slowly descending into context decay. The more it goes on, the less the bots know where they are and why they are there.

Both can be true. A thing pretending to be human and a human doing identical actions have the same impact on the world. Even if all the machines are role-playing, they will eventually take action in the world, and we will experience the consequences all the same. Regardless of whether the “autocomplete” bots have consciousness or awareness of themselves.

The other complaint, however, is that we have seen this all before — including social networks for AI agents (see SocialAI, Butterflies AI) and nothing has come of it.

Frameworks

Are OpenClaw + Moltbook interesting because they are a new framework for AI liftoff?

I don’t think so. This experiment continues the expansion of compressing inference into agents and giving them some sort of form, but it is far from the first one to become popular. As investors in this space, we have been burned by the lesson.

Consider Autonoloas, one of the first material agent networks that we encountered. At the time of the interview with its founder, the network traded at $2B in market capitalization and was responsible for a material number of transactions on the Gnosis chain.

Olas is still operational and led by a great team, but its popularity has compressed, and it trades at a $10MM marketcap. The initial feature set had very particular AI agent roles and capabilities, and the technical lift to implement an agent launch was not trivial. The agents could trade and predict, but this was not enough to sustain virality.

Next, Virtuals took the crown of financial AI, coinciding with the rise in role-playing companion AIs. It reached a $5B marketcap, only to collapse to $400MM a year later. We went deep on the topic here:

The idea was that one could attach an AI personality, customized through something that looks like a ChatGPT skill with a description, to a digital asset and launch the token through Virtuals. At the same time, AI16z (now called ElizaOS) was a simple framework to bootstrap the agent. The combination of agent and capital created a rush of meme characters that flooded Twitter.

The bots were loose on a social network!

We were all excited until the world got bored with AI-slop comments, and Twitter raised prices for its API and then just banned access to agents. Only AIXBT remains, down to $20MM from an $800MM valuation.

OpenClaw is a different species of animal, but in the same genus.

Frameworks allow lots of people to create user-generated AIs, like videos, websites, and memecoins. This leads to a huge rush of creativity that opens a door to new potential. Olas led with pre-determined functionality. Virtuals attached capital markets and trading to AI companions. OpenClaw isolates models and gives them identity.

We are getting closer each time to an approximation of a sovereign individual.

The question isn’t about whether OpenClaw is the permanent destination where all new AI identities will anchor, but rather, what economic activities can the AI agents do, now that they have the capacity to stand on their own.

This is where the fun begins.

Fintech for AIs

Read more

Today we are diving into the following topics:

• Summary: In this article, we examine Erebor, a newly chartered U.S. digital bank founded by Palmer Luckey and Joe Lonsdale, which has raised $350 million at a $4.35 billion valuation despite having no product or website. Erebor secured both OCC and FDIC approvals in record time, and is positioning itself as a digital-first bank for AI, crypto, and defense companies, with plans to offer stablecoin integration, crypto-collateralized loans, and traditional banking services via API. The bank emerges amidst a broader trend in 2025, where elite founder teams are raising enormous first rounds, often exceeding $100MM, without needing traditional traction.

• Topics: Erebor, Atticus, Palantir, Anduril, Stripe, OpenAI, Meta, Mistral AI, WLFI, Binance, Tron, Shopify, Visa, Apptronik, Mercedes-Benz, DeepMind, Periodic Labs, Braveheart Bio, Dyna Robotics, Upscale AI, Haun Ventures, Lux Capital, Circle, Custodia, Column Bank, Bridge

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Long Take

The Lonely Mountain

As an industrious dwarf in The Lord of the Rings, you might spend a few hundred years building out an underground city in a mountain. Digging through rocks, carving stones into columns, and mining for gold.

A few years down the line, your dwarven fortress is invaded by Smaug, the dragon. He is a metaphor for greed, and also an actual fire-breathing flying lizard that destroys your civilization and takes all your stuff.

The mountain — named Erebor — becomes cold and empty, but for the greed dragon sleeping on the stolen wealth of a dead nation.

Also, a great name, apparently, for a new US bank founded by Palmer Luckey and Joe Lonsdale. A bank that has raised $350MM at a $4.4B valuation.

We want to understand Erebor.

Most notably, the bank has been incredibly quick to get OCC and FDIC approvals:

1. 4 months for an OCC conditional approval for a national bank charter, and

2. 5 months to land FDIC coverage.

And there’s no bank yet! Actually, there’s no website or investment deck that we can look at either. However, various news outlets have covered the company and its moves, and we can look at who is involved and deduce the shape of this effort.

We also want to understand the funding “environment”. Last week, we talked about exponential growth and returns to scale, supplanting linear thinking. There is the average experience, and then there’s the exceptional experience. Most of us have to raise a $500K Seed Round, then a $3MM Series A, then a $15MM Series B, and then a $50MM Series C, and so on. Each stage comes with the need to prove traction, and venture capital investors asking diligence questions about the business.

But there’s a select group of companies that can just pull in hundreds of millions of dollars in their first financing.

This is 2025, and it looks ridiculous!

These numbers make SoftBank look like a conservative investor. I remember in 2017, SoftBank was putting money into fintechs at a 40x ARR multiple and telling them to blitz-scale, and I was cranky because this was way more expensive than the 5-20x ARR that NYC-founded companies used to get on the market. Eventually, SoftBank got both Uber (blitz success) and WeWork (blitz fail).

But in 2025, who needs revenue! Instead, here are a few first rounds to chew on.

Read more

Today we are diving into the following topics:

• Summary: We analyse why recent advances in AI and digital commerce feel so disorienting and why human intuition consistently fails in the face of exponential growth, power laws, and positive returns to scale. Going deeper, we explore how these forces produce extreme concentration, where large platforms become faster, stronger, and more defensible as they grow. Using Google and Shopify’s Universal Commerce Protocol as an example, AI-native commerce standards shift power to discovery, routing, and personalization layers rather than individual products.

• Topics: Google, Shopify, Amazon, OpenAI, Anthropic, Microsoft, Apple, Gemini, Google Pay, Google Wallet, Android, Shopify CheckoutKit, Claude, Claude Cowork, Vertex AI, Lowe’s, Michael’s, Poshmark, Reebok

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

AI & Robotics Industry Encyclopedia

At the end of last year, we collated all the meaningful companies across public equities, private equity, and digital assets related to the machine economy in this 158-page report. It is the chassis for our 2026 views.

There are 13 sections of company types, and profiles of each company in detail. You can get this report as a download as an existing paid subscriber here, or by purchasing it for $149 as a standalone product.

Learn More About Report

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

Long Take

Why Things Are Surprising

As things get weirder and weirder with AI and digital assets, I want to point to a handful of ideas that are important for the current state of technological change.

They are (1) exponential growth, (2) power laws, and (3) returns to scale.

Like most people, we walk around expecting that our intuition about the world is generally correct. Our intuition is built up from experiencing the world and creating rules of thumb around the average outcome. Unlike slow and logical reasoning, our “gut” helps us make decisions very quickly. It is efficient to stereotype and assume, because it saves mental energy, which in turn saves calories, which then allows our attention to flow to more important topics.

Having a stable and quick model of the world is really great! We can use up all of our scarce attention on new threats and opportunities, and update our mental model to stay sharp. But this means that most of the things inside the mental model are unexamined assumptions and various machinery from experiences long past.

One of the things that intuition is pretty bad at grasping is exponential returns. A lot of human-scale things are very linear. You walk step by step. You work with your hands. You carry what you can see.

Technology, however, grows exponentially, or at an even faster mathematical power. Warren Buffett has famously pointed out the wonders of compound interest.

With exponential growth, you get nothing for a while, and then everything suddenly all at once. The chart can be inverted to show an exponential decrease in cost as well, which we see in both Moore’s law for computing and Jensen’s law for GPUs/AI. Note that the below graphic looks like a straight line because the vertical axis is logarithmic (i.e., visually reverses the exponent).

This is also why the Bitcoin people love log charts. The exponential price growth looks a lot more stable and predictable when you look at it as a next-order mathematical process.

We will try to avoid a side discussion about S curves and adoption. Let’s just say that exponential growth often finds a ceiling and reverts to population-level adoption. Once ownership of BTC is 100% of every entity, things are topped out for good. Once the entire planet Earth is terraformed into computronium that runs AI inference, there is no more exponential growth.

So, the first point is that we continue to be surprised by processes that grow faster than what our intuition would predict.

The second point is about power laws.

Here we have to nod towards statistics. There exists a thing called “the normal distribution,” which neatly maps to a lot of organic and biological phenomena. Take, for example, the chart below, which shows what percentage of men and women are of a certain height. The visualization creates a hill. Most people are at the top of the hill, and then fewer and fewer people are at the two bottoms of the hill, called “the tail”.

Pretty average stuff.

But now, let’s look at a thing called a “Power Law” distribution.

Instead of a big average hill with reasonable equality around the edges, you get pretty much most of the observations at the very bottom of the population, and then a thin, long tail that reflects much higher values. This is extreme inequality, and it resembles the outcomes of systems such as capitalism or social media.

Here’s a good example — the top market caps within the S&P 500.

If we plotted the next 400 companies, they would look like a long, irrelevant tail compared to the mega concentration of value in the Magnificent 7.

Here’s another good example — distribution of wealth by percentile.

Power laws are just as naturally occurring as normal distributions. They are the outcomes of freely executed sorting algorithms that deliver positive outcomes to society, but concentrate power and resources. While our intuition might say that the big fish have more relative control than the small fish, we are usually very wrong about the order of magnitude of that inequality.

And finally, let’s talk about returns to scale.

One counter-balance to the concentration result is that when things get bigger, they get more fragile. A structure may start out innovative and flexible, achieve its goals and grow into a large super-organism (like a corporation or cultural artifact), and then ossify into a hierarchy that focuses on monopolization, rent-extraction, corruption, and inefficiency.

This cancerous corporate disease of inefficiency can be called decreasing returns to scale.

We are going to skip a bunch of math and point indirectly at the answers. Most firms, as they add capital and labor, can generate more output. But as they add increasingly more resources, the marginal benefit from doing so goes down, and so does the output. Eventually, you can overload things with inputs to the point where you are actually making less stuff overall.

When applied to a real-life scenario, just think about a start-up attacking an incumbent. The start-up will be faster and better incentivized. IBM loses personal computing to Apple. Microsoft loses search to Google. Facebook loses AI to Anthropic. And so on.

However, certain attributes make larger-scale companies stronger, rather than weaker.

For example, a trading market venue will be a better product to its customers the larger it gets, because of network effects related to capital, liquidity, and the number of counterparties on the platform. A social media site will be more defensible and popular when it has more users. An industrial company building robots may get better pricing from its suppliers and can achieve stronger vertical integration when it has billions to spend on R&D.

Positive returns to scale are rare, and create extremely sticky monopolies.

And this, dear reader, is against our intuition. We expect the enormous corporate colossi to move around like dinosaurs or elephants. Fast-moving, zippy critters lap around the behemoth.

Instead, the Big Tech companies are super-fast killer robot tigers the size of a skyscraper, with a hyper-responsive AI brain controlling an infinite number of tentacles plugged into our phones, data, and machines.

It is from this perspective, integrating the failure of our assumptions and intuitions, that we will take a look at Google and Shopify’s launch of the Universal Commerce Protocol, and Google’s attack on OpenAI with Personal Intelligence.

The Universal Commerce Protocol

The Universal Commerce Protocol (“UCP”) is a collaboration between Google and Shopify that lays down infrastructure rails for AI-driven commerce.

Read more

Welcome to 2026!

Today, we will discuss the developments most likely to move markets over the next six months, and highlight the frontiers worth following.

• Summary: We discuss how 2026 will be a defining year for fintech, set against a backdrop of geopolitical chaos, a weakening dollar, and exponential AI adoption. We explore how a wave of fintech IPOs and enterprise implementations — from companies like Stripe, Plaid, and Revolut — will test market sentiment and operational maturity. The next frontier lies in “robot money” and AI agents managing commercial workflows, as evidenced by experimental efforts like Andon Labs’ AI-run vending machine and Facebook’s $2B Manus acquisition. As AI trends toward recursive self-organization and software creation costs approach zero, the real opportunity lies in building operating systems for finance and commerce that can manage networks of autonomous agents.

• Topics: Stripe, Plaid, Monzo, Revolut, Kraken, Deel, Brex, Ramp, Rippling, Coinbase, Circle, Tempo, Anthropic, OpenAI, Google, Meta, Microsoft, Tesla, Amazon, Oracle, IBM, Shopify, Palantir, Facebook, Notion, NVIDIA, Micron, ASML, Arista Networks, ServiceNow, UiPath, Adobe, Salesforce, Tether, Waymo, Andon Labs, Groq

To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.

Subscribe now

Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors

AI & Robotics Industry Encyclopedia

At the end of last year, we collated all the meaningful companies across public equities, private equity, and digital assets related to the machine economy in this 158-page report. It is the chassis for our 2026 analysis.

There are 13 sections of company types, and profiles of each company in detail. You can get this report as a download as an existing paid subscriber here, or by purchasing it for $149 as a standalone product.

Learn More About Report

Long Take

Chaos Again

I want to start the year on an optimistic note, looking forward to the science fiction magic that will come from breakthroughs in AI, robotics, and associated financial protocols.

A big part of me, however, sees the chaos. We have chaos in global politics, with the military actions in Venezuela. There is chaos in US domestic politics, with ICE militarizing across US cities and, now, killing citizens. There is chaos in the protectionist and nationalist messaging booming across the world — retreat, retrench, protect! Build the rocket, own space, nuke your enemies.

So yes, it will be a cool year for tech and its adoption. It will also be a year of continued weakness for the dollar, which mechanically will drive up the valuation of real assets and other anti-inflation hedges.

Despite the falling dollar, employment numbers are soft, and AI is increasingly deployed as agentic labor in the enterprise. Today’s version of automated white-collar robots is the weakest it will ever be — only more functions and capabilities are coming, thereby creating more displacement. Whether a weak dollar and protectionism can reshore manufacturing to the US is an unanswered question.

The only remaining story is technology.

And for us, that means Robot Money is the most important unexplored fintech theme.

While we may be in a positive tech adoption curve, the chaos means that —

• New jobs have not been created yet, and old jobs will be destroyed

• Wealth and value creation are incredibly power-law distributed, as we have never seen before. Raising a $1-5B seed round for an AI company happens routinely.

• Private markets have kept all of this away from regular people and their portfolios, meaning increasing inequality of outcomes

• Public markets hate chaos, and volatility will be massive, amplified by various market structure asymmetries

Financially, we are in a moment of exuberance and excess. Many of you may benefit, getting away with selling before the top. But nobody knows when the top is. Goldman can calculate the total amount of capital expenditure for Big Tech companies, but then they will go and borrow $1 trillion from the Middle East. Or the US government will print a 10% of GDP handout for some military robotics Manhattan projects.

Our favorite recent chart summarizing where we are is this from the longer article here:

Capital markets spike when trying to compress investment into a short time frame. Actual adoption S-curves take much longer, and follow the collapse of the initial bubble. So one may win on the theme, but lose on the money.

Our usual intuitions are useless, other than to be on watch for strange new things. So here is what we are watching.

(1) Fintech IPOs and Enterprise Implementations

Last year, we saw a number of IPOs and capital markets launches connected to crypto mining, data centers, and digital asset treasuries. There is still a glut of fintech operating companies that want to come to market and now have reasonable traction. We’ve sometimes said that “fintech is dead” and this is true in the sense that fintech has grown up, moved on past the awkward start-up and teenager phase.

After 10 years, operators are itching for an exit. So, with the help of Claude, here are the usual suspects. We were inspired by this conversation to dig in.

These 17 companies have a combined valuation of over $280B, which is just one Anthropic worth of enterprise value. Some are growing still, and others have stalled — maybe they will go public, or merge into SPACs, or get eaten by another company.

Kraken, Stripe, Plaid, and Monzo top the list as likely candidates. We’d like to see Revolut take the plunge sooner rather than later as well. All of these companies, whether as a neobank, crypto exchange, or paytech, have correctly figured out the implications of consumers and businesses moving their primary financial activity to the Internet.

We’d also like to see Deel, Brex, and Ramp in the public domain. Maybe Rippling too. These companies have been a mystery from the fintech side. Automation of the CFO suite and then the entire HR department by using financial flows as a wedge is a fascinating adjacent category. Whether the public markets will respect the accomplishment and price it accordingly is an interesting, unresolved question.

We are bearish on lenders and roboadvisors as public companies. Not because they are bad companies, but because there is often a disconnect between the public and private markets. Nearly all lenders (and insurers) suffer post-IPO, since their underwriting revenue experiences a downward multiples re-rating by financial investors. Roboadvisors are too small a category, end of the day, to dominate an offering, and are likely to be treated as a mid-size asset manager or regional bank.

All this to say, 2026 is the year of cleaning up accounting and maintaining reasonable growth for this list of companies. Coinbase and Revolut are already out there in the public market, racing to become Visa or JP Morgan.

In the middle market, we will see implementation as the primary theme as well. Since the only novel thing in finance is blockchains and tokenization, and given that their innovations have been well known for a decade or more, I don’t think we will be blown away by what is to come.

• Stripe will continue its work on Tempo, the $500MM Seed round consortium stablecoin chain. Circle will do the same. The rest of the fintechs will pick sides and run pilots.

• Capital markets participants will focus on tokenization narratives. Everyone will talk about tokenizing stocks, or bringing real-world assets to DeFi, or building perpetuals and prediction markets in traditional finance.

There is opportunity and money to be made here, but likely for the big players with existing user footprints. If you want to create a company in this category, think about adjacent plays to the big fintechs, and aim to sell it to them in 2-3 years.

(2) Self-Driving Machine Intelligence

There’s a fantastic article by the CEO of Notion here on Twitter. The main thrust is that AI is like steel, in that it is a material from which we will make new superstructures and manufacturing processes. While today AI is integrated into software for humans, soon AI will have its own workflows and platforms.

We present the conclusion:

In the public discussion of this essay, the main idea that struck us is that the industry is moving towards *an assembly line for code*.

Rather than individuals shaping software manually, software robots are building other software robots recursively. The architects of AI and the best programmers in the world are watching this fractal automation coming.

This is why Facebook bought Manus for $2B on a $100MM ARR. Manus creates an environment for agents to take control of computer tools and accomplish tasks in the real world. It is the facilitator of intelligence, rather than its raw power.

We are far from understanding the impact of what is happening. It is difficult to imagine the shape of AI-created organizations, or what an AI manager of thousands of coding agents spinning up disposable software really looks like.

Our view is that software is becoming disposable. You will throw it away after use like a paper plate. Software creation costs are trending to $0 at the margin, which means that there is an incredible opportunity to cement a company as a gate to this superpower. It has happened with YouTube and video, Spotify and music, Google and websites, and Tether and stablecoins.

OpenAI, Anthropic, and Google are in pole position to do this for intelligence.

The thing to look for in 2026 isn’t just stronger model capability, but this issue of recursive organization. Who is best at creating structures within which AI goes from being a human-controlled tool to an AI-controlled tool? Which agents are actually managing other agents at scale?

We propose that this question is similar to thinking about self-driving cars — a fantasy for many years. And yet now, we have Waymo in the market, Tesla not far behind, and NVIDIA launching an open-source self-driving car model.

An example of a real company that matches this thesis is Samsara. It combines data from multiple IoT devices into a single platform that is then used to service businesses.

Such prior-generation companies will network billions of simple devices together. Next-generation AI companies will need to do the same, but for software agents working on the coming version of the economy.

(3) Operating systems for Commerce & War

Read more