Good morning & happy Wednesday! Today we’re diving deep into Anthropic, which stopped selling AI to Wall Street and started becoming Wall Street's OS (what the 10 finance AI agents actually do, why FactSet dropped 8.1% the same day, what the FIS infrastructure deal and $1.5B Blackstone/Goldman JV mean for distribution & what to expect next + bonus Guide on Claude Managed Agents, Claude Code Routines, & How to Build Your First AI Agent from Sratch inside), and Coinbase, which just did its 3rd layoff in 4 years and blamed it on AI (why it's actually a business model story, the real numbers behind the cuts, and what’s next + why now it’s a perfect time to Productize Yourself & Anthropic that just told AI founders exactly what to build in 2026). So let’s jump straight into the paradigm-defining stuff 🌶️

Anthropic stopped selling AI to Wall Street and started becoming Wall Street’s Operating System 🏦🧠

The BIG News 🔥 The most telling detail from Anthropic’s May 5 Finance Event wasn’t the 10 agent templates, the Moody’s integration, or Jamie Dimon sharing the stage with Dario Amodei. It was the fact that FactSet dropped 8.1% the same day.

The market read the announcement correctly. Anthropic didn’t ship another chatbot for bankers.

• It shipped a full AI agent platform that sits above every data provider, every Microsoft Office app, and every internal system a financial institution runs, orchestrating end-to-end workflows that used to require teams of junior analysts and weeks of calendar time.

• Ten production-grade AI agent templates covering everything from pitchbook assembly to month-end close to KYC screening.

• Native Excel, PowerPoint, and Word integration.

• Apache 2.0 on GitHub.

• Citadel, Goldman, BNY, Carlyle, and AIG are already deployed.

• And the real distribution play - an infrastructure partnership with FIS , which processes core banking for institutions handling roughly 12% of the global economy - means thousands of banks will get Claude’s agents without ever signing an Anthropic contract.

Combined with a $1.5B joint venture announced the day before with Blackstone, Goldman, and Hellman & Friedman, Anthropic now has three simultaneous paths into finance: (1) direct enterprise, (2) infrastructure embedding, and (3) a PE-backed services arm reaching hundreds of mid-market firms.

The architecture, the partnerships, and the market reaction all point to the same conclusion: the value in financial services is migrating from the data layer and the workflow tool layer to the intelligence layer above both.

Anthropic intends to own that layer.

Let’s unpack what it means for data incumbents, fintech startups building in the space, and the firms racing to adopt, or trying to figure out whether they’re already too late.

Read more

Good morning & happy Tuesday! Today we’re diving into AI giants OpenAI & Anthropic, which just outsourced their sales teams to private equity (breaking down their PE playbook, why it matters & what to expect next + bonus deep dives into OpenAI that wants to build the last App Store while also developing a personal AI CFO, & Anthropic’s AWS strategy), and a 3-year-old firm nobody’s heard of that just bought the world’s largest corporate travel company (why Long Lake Management is acquiring Amex GBT, and what it means for Ramp, Brex and Navan + bonus deep dives into American Express and why it skipped the Agent Protocol wars & Anthropic, which just told founders exactly what AI products to build in 2026). Let’s jump straight into the fascinating stuff 🌶️

OpenAI and Anthropic just outsourced their sales teams to private equity 🤑💸

The news 🗞️ The two leading AI labs stopped pretending software sells itself.

On the same day, OpenAI and Anthropic each finalized & signed joint ventures with the biggest names in private equity - TPG, Blackstone , Goldman Sachs , and Bain Capital - to physically embed their engineers inside thousands of portfolio companies.

The combined capital committed: over $5.5 billion. The model: Palantir’s forward-deployed playbook, scaled to the entire mid-market economy in one move.

One of these deals includes a 17.5% guaranteed annual return to investors, which is either a masterstroke or a ticking liability on top of $14 billion in projected losses. The other offers no guarantee at all, and that contrast tells you almost everything about where these two companies stand heading into their IPOs. For fintech operators and financial services investors, Goldman’s involvement on the Anthropic side opens a direct pipeline into wealth management, lending, and insurance portfolios that most startups spend years trying to reach. And that could be huge.

Let’s break down the deal structures, the real competitive implications for enterprise software & consulting, and the 3 specific things to watch over the next 12 months.

Read more

Subscribe now

On April 30, 2026, Anthropic published what most people read as a safety paper. It was, in fact, one of the most actionable product-market fit signals ever released by an AI lab.

Titled “How People Ask Claude for Personal Guidance,” the study analyzed 1 million Claude AI conversations and found that roughly 6% were people seeking personal advice - not information, not code, not summaries, but genuine life guidance on health, careers, relationships, money, parenting, legal rights, and more.

Across hundreds of millions of AI interactions, that translates to tens of millions of people using AI as a substitute for a doctor, financial advisor, therapist, lawyer, career coach, or relationship counselor - often because they cannot access or afford the professional alternative. Anthropic states this explicitly: users told Claude they turned to AI “precisely because they could not access or afford a professional.”

We already have proof that this data converts into venture-scale outcomes. Cal AI - an AI calorie tracker built by two high school students - solved one sub-topic from one domain in Anthropic’s taxonomy: “calories and macros for body composition.”

That single product grew to $40 million in revenue, was acquired by MyFitnessPal, and crossed $50 million in annualized revenue - with 7 employees and $0 in venture funding beyond the founders’ conviction. And health and wellness is just one of nine domains. Each one contains multiple sub-topics of equivalent or greater scale.

But the study is only half the signal.

Anthropic’s current job postings expose a parallel story that no one else is mapping: the company is now building vertical AI products in four domains - healthcare, financial services, legal, and life sciences - with dedicated engineering teams.

Every one of these builds targets enterprise customers, not consumers. That thus creates a precise line between where founders face platform risk and where the consumer layer is wide open.

This deep dive cross-references both signals - the guidance study and the careers page - to produce a domain-by-domain analysis covering all 9 personal guidance verticals.

• For each domain, we map Anthropic’s internal build status, identify funded startups already competing (with pitch decks where available), quantify the stakes and willingness to pay, and lay out a specific founder playbook.

• We also analyze the Claude Marketplace as a distribution channel, the Q1 2026 capital environment (including why VCs have stopped funding thin wrappers), and the cross-domain investment thesis linking Anthropic’s data to pricing, margins, and acquirer strategy.

• Plus, to make building easier, we’re also sharing How I built an AI operating system to run a startup with Claude 🤖, How to Turn ChatGPT Images 2.0 & Claude Design Into Your Chief Designer 🫟 (to build app prototypes, pitch decks, etc.), and The List of Top 100 Seed Investors 💸.

Let’s dive in.

The Framework: Why This Study Is a Product Roadmap

What Anthropic Actually Measured

Read more

If you’re not a subscriber, here’s what you missed this week:

1. He Got Fired From OpenAI at 22. Then He Turned $225M Into $5.5B by Betting on the AI Infrastructure No One Else Was Watching 💸 [Inside Leopold Aschenbrenner’s Situational Awareness Fund: the AGI investment thesis, the full portfolio breakdown, and the framework any investor, entrepreneur, or operator can steal right now]

2. Turn ChatGPT Images 2.0 & Claude Design Into Your Chief Designer 🫟 [How to build a Brand Identity, Design System, and Product Prototype in hours - without a designer. It’s the best 2026 workflow for FinTech & AI founders who need institutional-grade design fast]

3. Claude Code Routines: 8 Production Prompts, Real Costs, and Where They Break 🤖 [Copy-paste prompts for autonomous code review, deploy verification, and deal flow screening - with the failure modes you won’t find in Anthropic’s docs]

4. Stripe is building Visa for Machines 🤖💳 [two most important product launches from Stripe Sessions 2026, & how the fintech giant is building the full financial stack for autonomous AI + bonus deep dives into Stripe & how Google just won Agentic Commerce]

5. The Definitive Guide to Perplexity Computer 🖥️ [How Perplexity’s 19-model AI Agent works, what it costs, and whether it’s worth $200/month in 2026]

6. How to Build an AI Agent from Scratch (With Working Code) 🤖 [The design framework, working code, and hard-won lessons that take you from “I want to build an AI Agent” to a working one - in under 60 minutes]

7. Inside Revolut’s PRAGMA: The Foundation Model Trained on 40 Billion Banking Events 🧠 [Architecture, performance benchmarks vs. Stripe, Mastercard, and Visa, regulatory risks, and why PRAGMA may be the most consequential AI bet in consumer finance]

8. OpenAI doesn’t want to build a Phone. It wants to build the last App Store 😳📱 [why it’s not about killing the iPhone here but all about building the last App Store + bonus deep dive into OpenAI’s OS strategy & the quest to become the ultimate AI Super App]

9. Instacart’s co-founder built an AI Hedge Fund with no portfolio manager 🤖💰 [inside look into Abundance, what it indicates & what to expect next + bonus deep dive into Leopold Aschenbrenner’s Situational Awareness Fund, the most successful investor on Wall Street right now & pitch deck teardown of AI Family Office inside]

10. Claude Cowork Commands, Scheduled Tasks & Automation Workflows: The Operator’s Playbook 🤖 [Every real slash command, the exact prompts that run deal flow and investor updates on autopilot, and the failure modes that will cost you hours if you don’t know them first]

11. Turn Claude Cowork Into Your Personal COO 🧠 [Most people are using the most powerful AI agent ever released to answer questions. Here’s the framework that turns Claude Cowork into an operator that runs your work while you sleep]

12. The One-Person Unicorn 🦄 [I built an AI operating system to run a startup with Claude]

As for today, here are the 2 fascinating FinTech stories that are changing the world of financial technology as we know it. This was yet another wild week in the financial technology space, so make sure to check all the above stories.

Google’s UCP just won Agentic Commerce. Stripe, Amazon, and Microsoft walked away from the only alternative 🤖💳

The news 🗞️ Every major player in agentic commerce just picked a side, and they all picked the same one.

Last week, Amazon, Microsoft, Meta, Salesforce, and Stripe joined the Universal Commerce Protocol’s Tech Council, the open standard governing how AI agents discover, cart, and purchase products on behalf of consumers. They join the founding members Google, Shopify, Etsy, Target, and Wayfair.

The protocol war for agentic commerce is over before most people realized it started.

The defection that tells the whole story is Stripe’s, which co-built the rival Agentic Commerce Protocol with OpenAI just seven months ago. That bet failed publicly. Now Stripe sits on the tech council of the standard it was supposed to compete against.

Let’s break down what Google actually gains here, why the 1.18% conversion problem haunts the entire category, and the payments compression signal every fintech investor should be modeling now.

More on this 👉 Stripe co-developed the Agentic Commerce Protocol with OpenAI to power Instant Checkout inside ChatGPT. The thesis was compelling: let agents handle the entire purchase flow natively, no merchant redirect required. The results were not. Walmart’s tests showed roughly 1.18% conversion rates and 77% cart abandonment. Outch 😳

OpenAI thus pulled back Instant Checkout in March 2026 and pivoted to merchant-controlled flows. Stripe read the room and walked across the aisle.

For Google, this is a masterclass in platform extension. As a UCP founding member, Google has positioned itself to own the protocol layer for agent-driven commerce, the same way it owns the discovery layer for human-driven commerce. Search was the chokepoint of the browser era. UCP could be the chokepoint of the agent era.

ICYMI: The Android of Commerce - How Google Is Building the Interface Between AI & Money 🤖💸 [why recently introduced WebMCP is a game-changer, how it stacks perfectly into Google’s Android of Commerce playbook, things worth watching, what’s next for FinTechs/Banks/Payments companies, etc. + bonus deep dives into Google’s UCP, how I built an AI OS to run startup with Claude, and how AI is eating software today]

The difference is that this time, Google got the whole industry to help build and legitimize the rails rather than fighting antitrust battles over proprietary ones. Sundar Pichai posting about “extraordinary progress” wasn’t a celebration. It was a flag planted 😎

Zoom out 🔎 The real question now isn’t adoption. It’s conversion.

UCP has every relevant company at the table, but as Shopify’s Ilya Grigorik noted, a standard’s success is measured by what happens after adoption. Can agent-mediated purchases actually convert better than a human clicking “buy now”? Nobody has proven that yet.

Most importantly, the 1.18% number from ACP’s failure hangs over the entire category, not just OpenAI.

THE TAKEAWAY ✈️

What’s next? 🤔 Looking ahead, watch two things. First and foremost, whether Google embeds UCP primitives directly into Gemini’s agent infrastructure, giving it a structural advantage over every other AI assistant in commerce. Second, whether payments economics change. If agents commoditize discovery and checkout into a protocol call, interchange and affiliate revenue models face compression. Stripe joined UCP because it had to. But the protocol that makes agent commerce frictionless could also make payments a thinner business. That’s the tension nobody on the Tech Council is talking about publicly, and every fintech investor should be modeling privately.

ICYMI: American Express skipped the Agent Protocol wars and bought the risk instead 🤖🛡️ [what The Amex Agentic Commerce Experiences (ACE) is all about, why it matters & what to expect next + bonus dives into Shopify’s Agents, AI Agents now opening bank accounts & how to build your first AI agent from scratch inside]

Revolut earned £1.3B last year without really being a bank. JPMorgan and UBS just mapped what comes next 😳🏦

Following the money 💸 Two of the world’s largest banks recently published extensive research on Revolut, and the portrait they’ve drawn deserves attention from anyone in fintech or finance.

Let’s take a look at this.

More on this 👉 J.P. Morgan initiated coverage with a 54-page report on April 23, the same day UBS released its latest BankTech Bulletin dedicated to Revolut’s FY25 results.

Together, they offer the most detailed institutional analysis yet of a company now valued at $75B, with the Financial Times reporting internal IPO discussions at $150-200B and a potential 2H26 secondary expected above $100B. What makes these reports worth reading closely isn’t the growth story everyone already knows. It’s the structural oddity both banks circle around without quite resolving: Revolut is printing 35% return on equity with a 6% loan-to-deposit ratio. It barely lends money.

Zoom out 🔎 We can remember that Revolut’s revenue hit £4.5B in FY25, up 46%. Net profit reached £1.3B, up 65%. The cost-income ratio compressed to 51% even as marketing spend rose 47% to £529M and the company pushed into Mexico, Colombia, and over a dozen new product categories.

Those are elite numbers. But net interest income accounts for just 24% of total revenue. At Monzo, that figure is 66%. At Starling, 87%. UK incumbents average 63%. Revolut sits on £36B in deposits and lends out £2.2B. Card interchange, subscriptions, FX, and crypto trading fees do the real work. Subscription revenue alone grew 67% to £708M. UBS pegs all-in impairments at just 0.2% of financial assets.

J.P. Morgan thus estimates that shifting £1B from cash into loans could add £72M in interest income, and capturing 1-2% of UK consumer lending alone could generate £0.7-1.4B in incremental NII.

The UK banking licence, fully unrestricted since March 2026, makes this possible for the first time.

THE TAKEAWAY ✈️

What’s next? 🤔 All in all, both reports converge on the same core tension, and it’s the thing to watch over the next 18 months. Revolut’s valuation prices in a future where it successfully scales lending across dozens of markets. But through-the-cycle underwriting remains completely untested at scale. Every percentage point the loan-to-deposit ratio climbs brings Revolut closer to traditional bank economics, and potentially closer to traditional bank multiples of 1-3x book rather than the current 12x. Three specific triggers will tell you which direction this breaks: (1) the pace of UK lending growth post-licence, (2) credit performance in the first European downturn Revolut faces as a real lender, and (3) whether the U.S. bank charter application (filed March 2026) gets approved. If Revolut can scale credit without blowing up its loss rates, the path to a $150B+ IPO is credible. If not, the market will reprice this as a very good payments company, not a generational bank franchise.

ICYMI: Inside Revolut’s PRAGMA: The Foundation Model Trained on 40 Billion Banking Events 🧠 [Architecture, performance benchmarks vs. Stripe, Mastercard, and Visa, regulatory risks, and why PRAGMA may be the most consequential AI bet in consumer finance]

Revolut is the most profitable FinTech on Earth, and it hasn’t even started lending yet 🤯📈 [unpacking the fee-income fortress that makes Revolut structurally different from every competitor, the lending optionality coiled inside that balance sheet, the regulatory dominoes now falling in sequence, and whether $75B is expensive or whether the market is about to find out it was way too cheap + more bonus reads on Revolut, how it’s leveraging AI & bonus deep dives into the latest financials of Nubank, SoFi, Robinhood, and Coinbase inside]

🧠 What else I’m watching

• Gemini Enables AI Trading Bots 🤖 Gemini has launched Agentic Trading, allowing users to connect AI models like Claude and ChatGPT to their trading accounts via the MCP open standard. This first-of-its-kind tool enables AI bots to autonomously monitor markets, place and execute trades, and manage risk based on predefined strategies - directly through a regulated US exchange. The platform introduces modular “Trading Skills” like finding spreads and retrieving historical data, with more functions planned. The move aligns with the broader agentic AI trend, where AI bots are increasingly integrated into digital services. ICYMI:

• Goldman Sachs Blocks Claude in HK 🚫 Goldman Sachs has restricted access to Anthropic’s Claude AI models for its Hong Kong bankers, following a strict interpretation of its contract with Anthropic. While employees retain access to other models like ChatGPT and Gemini, the ban on Claude aligns with rising US-China AI tensions, after the US government warned of Chinese firms allegedly stealing AI technology. Anthropic’s models, including Claude, are not officially available in mainland China, and Hong Kong is not listed as an accessible market on Anthropic’s website. The move also follows concerns about the safety of Anthropic’s latest model, Mythos, which can reportedly expose cybersecurity vulnerabilities faster than humans. ICYMI:

• Citi Welcomes AI Wealth Assistant 🤖 Citi has partnered with Google to launch Citi Sky, an AI-powered avatar designed to enhance its Citi Wealth client service. Developed using Google Cloud’s Gemini Enterprise Agent platform and Google DeepMind’s real-time avatar technology, Citi Sky will integrate into Citi Wealth platforms in the US, working alongside financial advisors. Initially, it will assist Citigold clients with tasks like CD maturity guidance and market insights from Citi’s chief investment office. Citi Sky supports natural voice conversations in English and Spanish, with more languages planned. ICYMI:

💸 Following the Money

• Philippines-based banking and lending platform Salmon Group has raised $60M in equity financing and another $40M in public bonds.

• New York-based cross-asset 24/7 trading platform Liquid has raised $18M in a seed funding round co-led by Neo and Left Lane Capital.

• MoonPay buys crypto security specialist Sodot.

👋 That’s it for today! Thank you for reading, and have a relaxing Sunday! And if you enjoyed this newsletter, invite your friends and colleagues to sign up:

Subscribe now

Subscribe now

A full-time Chief Operating Officer costs $300,000 to $500,000 a year. Claude Cowork costs $20 a month. That comparison has been made before - we made it ourselves in the AI COO Guide (if you haven’t read it, start there; it covers the mental model, context file setup, and 10 foundational prompts that turn Cowork into a delegation engine).

But having the right mental model is only half the problem.

The other half is the operational machinery: which commands actually exist, how the automation layer works, and where it breaks, what Scheduled Tasks can and can’t do versus Routines, and the specific workflows that compound for founders managing deal flow, operators scaling teams, and investors processing pipelines.

That’s what this piece covers - the systems layer.

I’ve spent weeks pressure-testing every feature against Anthropic’s official documentation and real-world runs since Cowork went generally available on April 9, 2026. Therefore, below is the complete command reference, the automation architecture explained with the precision it deserves, five fully engineered workflow prompts built for high-ROI business use cases, the token cost math by workflow type, and every failure mode I’ve found with the workaround that makes it manageable.

Here’s a preview of what’s inside:

• The verified command reference. Every real Cowork slash command, what it does, and the ones you’ve seen shared that don’t exist.

• Scheduled Tasks vs. Routines. Two different automation systems. Two different products. One runs on your desktop (& dies when your laptop closes). The other runs in the cloud (but belongs to Claude Code, not Cowork). The distinction determines whether your morning briefing is waiting for you or was never generated.

• Five workflow prompts engineered for founders, operators, and investors. A deal intake pipeline that scored 300 pitch decks and cut first-pass review from 60 minutes to 2. A call-to-action system that auto-generates debriefs and follow-up emails from your Fathom or Fireflies recordings. An investor update writer that drafts in your voice with one honest challenge baked in. An SOP machine that converts the process knowledge in your head into documentation a new hire can follow. And a CRM pipeline hygiene workflow that catches stale deals, stage mismatches, and concentration risk every Friday before the week ends.

• The failure modes & cost math. Token consumption by workflow type (a 10-PDF synthesis burns 80,000–120,000 tokens - that’s a Max Plan job). The seven ways Cowork fails and the specific workaround for each. Which plan you actually need based on what you’re running.

Let’s get into it.

Every Cowork Slash Command That Actually Exists

Read more

Good morning & happy Thursday! Today’s issue is really hot as we’re diving into Stripe, which is now building Visa for Machines (two most important product launches from Stripe Sessions 2026, & how the fintech giant is building the full financial stack for autonomous AI + bonus deep dives into Stripe & how Google just won Agentic Commerce), and Revolut, which is now being covered by Wall Street’s biggest banks (key takeaways from JPMorgan & UBS reports on the most valuable FinTech in Europe, what’s interesting & what to watch for next + bonus deep dive into Revolut’s latest 2025 financials & how and why it built PRAGMA, the foundational AI model for money). Let’s jump straight into the interesting stuff 🌶️

Stripe is building Visa for Machines 🤖💳

The news 🗞️ For years, AI agents could do everything except the thing that actually matters in commerce: spend money. As of this week, Stripe fixed that. Twice over, in two very different ways, aimed at two very different futures.

One lets your AI shopping bot tap your wallet with biometric approval. The other lets machines stream sub-cent stablecoin payments to each other, with no human-in-the-loop at all. Put them together, and you’re looking at the most ambitious play in fintech right now: a full financial stack for autonomous AI, built by the company that already processes a trillion dollars a year in human payments.

The playbook is clear. What’s less clear is whether anyone can catch up.

Let’s unpack this.

Read more

Subscribe now

My team used to spend 4 hours every Monday morning triaging the weekend’s backlog. Labeling issues, pinging owners, checking whether Friday’s deploy held. Every Monday. Every week.

Then we set up two Claude Code Routines. Now we walk into a labeled queue, a PR digest, and a Slack message confirming the deploy is clean. Four recovered hours per week, every week, without a single person doing the work.

Claude Code Routines - launched April 14, 2026 in research preview - let you define a prompt, point it at a GitHub repository, attach a trigger (a cron schedule, an API webhook, or a GitHub event like pull_request.opened), and let Claude execute it autonomously on Anthropic’s cloud infrastructure.

No open terminal. No running laptop. No cron daemon to babysit. The agent picks up the work and delivers the output while you sleep 🤖

Below, we walk through eight production-ready routines with full copy-paste prompts:

• Automated PR code review that leaves inline comments on security, performance, and style issues.

• Deploy verification that posts a go/no-go verdict to Slack within minutes of a production push.

• Weekly docs drift detection that catches every stale API reference before your users do.

• A morning meeting prep agent that briefs you on every external attendee by 7 am.

• An expansion signal scanner that flags enterprise upsell signals in your support queue.

• A deal flow screener agent for investors processing 100+ inbound decks a week.

• And a portfolio health pulse that surfaces engineering velocity drops across your companies before board meetings do.

Each routine includes the exact prompt, the connectors required, what goes wrong in practice, and the guardrail line that fixes it.

We also cover the parts most people skip: the daily run caps that will catch you off guard (Pro plans max out at 5 routine runs per day), the security reality of every routine running under your personal GitHub identity, prompt injection risks when routines ingest external data like support tickets and customer emails, the four failure modes that will burn you if you don’t design for them, and the cost math that determines whether batching five tasks into one meta-routine is worth the complexity.

If your team is evaluating Claude Code Routines for production use, this is the piece that saves you the first month of mistakes.

To make the most of Claude AI, we’re also attaching The Ultimate Guide to Claude Managed Agents 🤖, and The Complete Founder’s Guide to Claude Design 🎨 inside. It has everything to make AI your most powerful leverage.

What Routines Actually Are

Read more

Good morning & happy Tuesday! Today, we’re diving deep into Google, which just won the Agentic Commerce race (why Stripe ditched their own protocol to join UCP, what Google gains here, & what to expect next + bonus deep dive into how Google is building the Android of Commerce & AmEx, and why it skipped the agent protocol wars altogether), OpenAI that’s now building an AI-first phone (why it’s not about killing the iPhone here but all about building the last App Store + bonus deep dive into OpenAI’s OS strategy & the quest to become the ultimate AI Super App), and Instacart’s co-founder Apoorva Mehta who built an AI hedge fund (inside look into Abundance, what it indicates & what to expect next + bonus deep dive into Leopold Aschenbrenner’s Situational Awareness Fund, the most successful investor on Wall Street right now & pitch deck teardown of AI Family Office inside). Let’s jump straight into the fascinating stuff 🌶️

Google’s UCP just won Agentic Commerce. Stripe, Amazon, and Microsoft walked away from the only alternative 🤖💳

The news 🗞️ Every major player in agentic commerce just picked a side, and they all picked the same one.

Last week, Amazon, Microsoft, Meta, Salesforce, and Stripe joined the Universal Commerce Protocol’s Tech Council, the open standard governing how AI agents discover, cart, and purchase products on behalf of consumers. They join the founding members Google, Shopify, Etsy, Target, and Wayfair.

The protocol war for agentic commerce is over before most people realized it started.

The defection that tells the whole story is Stripe’s, which co-built the rival Agentic Commerce Protocol with OpenAI just seven months ago. That bet failed publicly. Now Stripe sits on the tech council of the standard it was supposed to compete against.

Let’s break down what Google actually gains here, why the 1.18% conversion problem haunts the entire category, and the payments compression signal every fintech investor should be modeling now.

Read more

Subscribe now

Leopold Aschenbrenner is either the most important investor no one on Wall Street has heard of, or the most dangerous kind of genius: one who might actually be right. In the twelve months following the launch of his hedge fund, Situational Awareness LP, in September 2024, he grew its disclosed U.S. equity exposure from roughly $225 million to $5.52 billion 😳

He did it without owning a single share of Nvidia, Microsoft, Amazon, Google, or Meta. His largest position was in a fuel-cell company. His second-largest was call options on a GPU cloud startup most investors had never heard of. His third-largest was call options on a chipmaker Wall Street had largely written off.

This is not a story about luck. It is a story about what happens when a mind trained in growth economics, existential risk theory, and AI scaling laws is given access to public markets and the conviction to act on conclusions everyone else considers too extreme to take seriously.

The Man Behind the Thesis

Leopold Aschenbrenner didn’t set out to run a hedge fund. He set out to understand how the world ends, and then figure out how to invest in the companies that would prevent it.

Born in Germany around 2001 to two medical doctors, he enrolled at Columbia University at fifteen. In 2021, at nineteen, he graduated as valedictorian with a double major in economics and mathematics-statistics, the same year he won a grant from Tyler Cowen’s Emergent Ventures. He did research at Oxford’s Global Priorities Institute alongside Philip Trammell, co-authoring a growth-economics paper that would later become the philosophical licence for his entire fund.

He interned at the FTX Future Fund, the philanthropic arm of Sam Bankman-Fried’s empire, and watched from the inside as one of the most spectacular coordination failures in recent financial history imploded. He resigned before the collapse. He then joined OpenAI’s Superalignment team in 2023, co-authored a paper on AI safety presented at a leading machine learning conference, and wrote an internal memo to the OpenAI board warning about Chinese espionage and the company’s security posture. OpenAI fired him in April 2024.

Two months later, he published Situational Awareness: The Decade Ahead, a 165-page essay series making a single sweeping claim: AGI is coming by 2027, the world is not ready, and the companies building the physical infrastructure to make it possible are the most undervalued assets in financial markets.

Ivanka Trump shared it. It went viral.

Within months, he raised anchor capital from Patrick and John Collison (Stripe co-founders), Daniel Gross, and Nat Friedman. One anchor LP told Fortune that Aschenbrenner pitched the fund by rejecting a venture-capital structure: AGI was going to be so consequential to the global economy that the only way to fully capitalise on it was to express investment ideas in the most liquid markets in the world. Public markets over private, because public markets scale with conviction and re-price faster.

Twelve months later, the fund’s 13F disclosed $5.52 billion in U.S. equity positions across 29 holdings. Let that sink in.

The Core Thesis

The One-Sentence Version: AGI is a 2027 event; superintelligence is a 2028–2030 event; the markets have not priced in either the exponential trendlines or the industrial mobilisation they require. Therefore: buy the bottlenecks of the buildout (power, compute, memory, optics, fabs) and short what AI eats.

What This Deep Dive Covers & Why It Matters

Most coverage of Leopold Aschenbrenner focuses on the headline number - 22-year-old, fired, now manages billions - and stops there. That framing is interesting but useless.

What actually matters is the architecture of thinking that generated those returns: the intellectual framework that turned a philosophical bet about artificial intelligence into one of the most concentrated, contrarian, and precisely sequenced portfolios in the hedge fund industry.

This deep dive goes several layers deeper:

• The academic paper almost no one cites - a plain-English explanation of the Trammell-Aschenbrenner paper Existential Risk and Growth, co-authored with Stanford’s Digital Economy Lab, and why its core result (that the risk-minimising rate of technological growth is typically positive and often high) is the philosophical licence for the entire portfolio. It is the direct intellectual foundation for every position the fund has ever taken.

• The 165-page manifesto decoded - the five core claims of Situational Awareness: The Decade Ahead, including the precise “Counting the OOMs” framework Aschenbrenner uses to forecast AGI by 2027, the three compounding vectors (compute, algorithmic efficiency, and “unhobbling”), and the specific test-time compute insight that explains the transition from chatbot to drop-in remote worker.

• Quarter-by-quarter portfolio evolution - five quarters of 13F data dissected, from the first $254.8M, six-holding filing in Q4 2024 to the $5.52B, 29-holding book filed in February 2026. When did he pivot from semiconductors to power to Bitcoin miners? Why did he simultaneously short the entire semiconductor sector while remaining long Intel and Broadcom? And the DeepSeek moment in January 2025 - when everyone else panicked, and he bought.

• The full portfolio breakdown - every position mapped to its thesis layer (electrons, compute, silicon, optics, memory, shorts), the activist 13D stake in Core Scientific, the Intel call-options bet decoded, the Bloom Energy fuel-cell play that Oracle just validated at 2.8 gigawatts, and the two private market moves the 13F doesn’t show.

• A live portfolio tracker - a tool I built with Perplexity Computer to track Situational Awareness LP’s positions in real time as new 13F filings drop, along with the whole prompt to recreate it yourself.

• The seven-rule reusable playbook - the complete framework, distilled from the thesis, the filings, and all available transcripts, that any investor, founder, or operator can apply to any technology S-curve, not just AI. The same logic that identified power companies and Bitcoin miners as the 2024–2025 winners can be applied to find the 2026–2028 equivalents - in robotics, in synthetic biology, in the physical supply chains of whatever comes after large language models.

• The honest risk register - six specific ways this thesis breaks down (including the 14:1 implied leverage ratio most coverage never mentions), and the six signposts you should be watching to know whether it’s still working.

Every major technology wave has a set of physical constraints that are obvious in retrospect but invisible to most market participants in real time. Aschenbrenner made them visible - and bet on them before the crowd arrived.

Understanding exactly how he did that is worth more than any single stock tip.

The Five Pillars

Read more

If you’re not a subscriber, here’s what you missed this week:

1. Turn ChatGPT Images 2.0 & Claude Design Into Your Chief Designer 🫟 [How to build a Brand Identity, Design System, and Product Prototype in hours - without a designer. It’s the best 2026 workflow for FinTech & AI founders who need institutional-grade design fast]

2. The Ultimate Guide to Claude Managed Agents 🤖 [How Anthropic’s new managed runtime for AI agents works, what it costs, when to use it, and how Notion, Rakuten, Asana, Sentry, and Atlassian are deploying it in production]

3. The Definitive Guide to Perplexity Computer 🖥️ [How Perplexity’s 19-model AI Agent works, what it costs, and whether it’s worth $200/month in 2026]

4. How to Build an AI Agent from Scratch (With Working Code) 🤖 [The design framework, working code, and hard-won lessons that take you from “I want to build an AI Agent” to a working one - in under 60 minutes]

5. Inside Revolut’s PRAGMA: The Foundation Model Trained on 40 Billion Banking Events 🧠 [Architecture, performance benchmarks vs. Stripe, Mastercard, and Visa, regulatory risks, and why PRAGMA may be the most consequential AI bet in consumer finance]

6. Coinbase’s App Store for AI Agents is a float play dressed as a protocol 👀📈 [why it’s really a float play for USDC & Base dressed up as a protocol, why it matters for Coinbase & what’s next + bonus deep dive into Coinbase’s latest financials & a full guide on building your first AI Agent From Scratch]

7. AngelList’s retail VC fund charges 2.5% to let you buy what your Uber driver already knows about 🥲📉 [why it’s not that exciting as it seems, what’s really under the hood here & what to watch for next + bonus list of recently funded AI startups, their pitch decks and what they tell us about the market & top 10 AI startups to watch in 2026 & their decks inside]

8. 10 AI Prompt Skills That Actually Change How ChatGPT, Claude, and Gemini Respond 🤖 [Stop telling AI to “write clearly.” Give it real frameworks with measurable constraints and watch the output transform]

9. The File That Turns Claude Code Into Your Best Engineer 🧠 [A complete guide to CLAUDE.md, rules, skills, subagents, and hooks - everything you need to make Claude Code your 10X Engineer]

10. Turn Claude Cowork Into Your Personal COO 🧠 [Most people are using the most powerful AI agent ever released to answer questions. Here’s the framework that turns Claude Cowork into an operator that runs your work while you sleep]

11. The One-Person Unicorn 🦄 [I built an AI operating system to run a startup with Claude]

12. Claude Design Just Made Design File Optional. Here’s the Complete Founder’s Guide 🎨 [Claude Design setup, 8 workflows with real prompts, the 7-step logo design flow, a full set of master prompts for rigorous design work, and what Figma’s collapse means for your next launch]

As for today, here are the 3 incredible FinTech stories that are transforming the world of financial technology as we know it. This was yet another wild week in the financial technology space, so make sure to check all the above stories.

Before we dive in, a quick word from our sponsor.

Watch the developer conference of the year. On demand on Salesforce+

Agentic AI is changing the game and Agentforce is leading the way. Watch TDX on demand to explore dozens of sessions covering the latest innovations across Agentforce, Data 360, the core platform, vibe coding, Slack, and more. All free on Salesforce+.

Watch TDX on Salesforce+ to:

- Get roadmap insights from the leaders shaping what’s next

- Access broadcast-only moments and exclusive interviews

It all starts with the main keynote, where you’ll experience the future of software and learn how to build it. Watch it now and catch every moment at your own pace.

Watch Now.

eToro stopped being just a broker, and nobody noticed 😳💸

The news 🗞️ One of the most consequential moves in retail finance this month wasn’t a rate decision or an IPO filing. It was eToro quietly shipping an App Store.

Let’s unpack this.

More on this 👉 WealthTech eToro just opened a marketplace where anyone - developers, quants, Pro Investors, or non-technical traders using AI “vibe coding”- can build and publish apps that plug directly into live trading infrastructure.

Algorithmic execution, social analytics, portfolio rebalancing, AI-managed sub-accounts: all accessible via API, MCP server, or natural-language prompts.

By launch day, 800+ Pro Investors had already built over 1,000 tools. Users install with one click through SSO. No separate logins, no data migration.

Zoom out 🔎 The playbook is obvious because it’s proven. Apple, Salesforce, Shopify all made the same bet: the platform that attracts the most builders captures the most users and becomes nearly impossible to leave. eToro is running that play against a competitive set - Robinhood, Interactive Brokers, Schwab - that either lacks a marketplace entirely or serves one aimed at professionals.

IBKR’s Investor’s Marketplace lists 1,500+ services but targets experienced traders. Bloomberg’s Open API costs $24,000/year per terminal. eToro is handing equivalent capabilities to 40 million registered users for free, then letting the community iterate faster than any internal product team could.

The timing is also tighter than it looks. Agent Portfolios launched in March (autonomous AI trading with $200 minimums, new registrations up 30% in Q1). Tori, eToro’s AI companion, just got Grok 4.2 integration and persistent memory. Then there’s the $70 million Zengo acquisition that gives eToro self-custodial wallet infrastructure for tokenized assets and prediction markets.

The App Store is the connective layer that turns these from features into a platform. And platforms usually win.

THE TAKEAWAY ✈️

What’s next? 🤔 Looking ahead, here’s what you should watch. First and foremost, the financial impact is back-loaded: retention and trading volume compound slowly, but the lock-in mechanics are real. Every installed app, every configured agent strategy, every follower relationship makes switching more painful. If eToro captures even modest developer energy, it creates a data flywheel - millions of live AI strategies generating proprietary behavioral datasets - that no competitor can replicate by copying the product. The risk is obviously curation. A flood of poorly built tools that lose people money turns the App Store from a moat into a liability, and regulators in the US and EU will be paying attention the moment AI agents start executing trades at scale. eToro’s current legal framing as “neutral technology infrastructure” is clever positioning, but it won’t survive contact with the first high-profile retail loss driven by a vibe-coded algo. Thus, the real question here isn’t whether eToro can launch an app store. It’s whether brokerage competitors can respond before the developer lock-in sets. Interesting times ahead 👀

ICYMI: eToro, or social investing’s hidden gem trading at a discount 👀📉

Cathie Wood bought Kalshi before she started using it 💸📊

The news 🗞️ The recent ARK’s Kalshi deal got mostly reported as prediction markets going institutional. But the sequence is the whole story here.

Let’s unpack this.

More on this 👉 In December 2025, ARK joined Kalshi’s $1 billion Series E at an $11 billion valuation, alongside Paradigm, Sequoia, and a16z. In March 2026, ARK became the flagship institutional customer, using a new “market request pipeline” to commission custom contracts on non-farm payrolls, deficit-to-GDP, and company KPIs in genomics and AI. Four months between buying the equity and validating the product publicly.

Press coverage separates the two events. It shouldn’t.

We must remember that five weeks before ARK’s announcement, Fed economists published a paper finding Kalshi outperformed Bloomberg consensus on CPI and held a perfect pre-FOMC forecast record against Fed Funds futures.

That’s the legitimizing event. Kalshi went from $2 billion in July to $5 billion in October to $11 billion in December, repriced not as a betting venue but as financial data infrastructure. Nick Grous calling the prices “purest expressions of risk” reads differently when ARK owns the cap table.

Zoom out 🔎 The ecosystem loop is perhaps the most interesting part. ARK’s public usage validates Kalshi to other institutions, deepens liquidity, expands TAM beyond retail into the full research stack, and marks up ARK’s own stake.

It’s classic venture-portfolio play run through a publicly disclosed asset manager. Nothing improper about it, but the “research partnership” framing obscures that ARK has two reasons to want this to work 😎

THE TAKEAWAY ✈️

What’s next? 🤔 Looking ahead, most people focus on whether sell-side research or Bloomberg gets displaced. The more important question though is compliance. Banks have started applying insider trading rules to prediction market positions, and the CFTC issued an enforcement advisory in February citing nonpublic information misuse on Kalshi. Every PM who wants to trade a company-KPI contract now has an MNPI question their compliance desk hasn’t answered. That’s the real friction, not pipeline infrastructure. Zooming out, watch ICE’s $2 billion Polymarket stake too; NYSE’s parent sees this as infrastructure worth owning, not a data feed worth renting. Kalshi’s moat has to become institutional integration, because regulatory legitimacy is about to commoditize.

ICYMI:

Robinhood’s OpenAI bet is really a bet on itself 🤖💰

Following the money 💸 The walls between retail investors and private markets didn’t come down last week. The SEC knocked them down months ago. Robinhood just walked through first.

Let’s take a look at this.

More on this 👉 Robinhood Ventures Fund I just closed a $75 million purchase of OpenAI common stock - actual equity, not the synthetic tokens that drew OpenAI’s public rebuke last summer. RVI shares jumped as much as 20% on the news. The fund, which listed in March at $25 per share and promptly sank to $22, suddenly had a reason to exist in the eyes of the market.

The backstory matters here.

In June 2025, Robinhood launched tokenized OpenAI shares for European users. OpenAI’s response on X was blunt: these aren’t our shares, we didn’t approve this, be careful. Ten months later, Robinhood holds real OpenAI common stock inside a registered fund with full regulatory blessing. That’s a remarkable rehabilitation of a relationship that started with a cease-and-desist tone.

Zoom out 🔎 But the OpenAI position is just the shiny object here. The deeper story is structural. Last year, the SEC dropped its 15% cap on closed-end funds’ private market allocations and eliminated accredited investor requirements for these vehicles. Robinhood had a $658 million fund ready to deploy the moment the door opened. The fund charges a 2% management fee on net assets, no performance fee. For a company whose revenue has historically swung with trading volumes, that’s a quieter, stickier revenue stream - and one that scales with every marquee name added to the portfolio.

OpenAI, valued at $852 billion after its $122 billion March round, is expected to file for an IPO targeting $1 trillion as early as late 2026. RVI’s stated plan is to hold through IPO and beyond. If that works, the fund becomes the proof of concept for a product category every brokerage will need to match.

THE TAKEAWAY ✈️

What’s next? 🤔 Looking ahead, here’s what to watch. NAV premium/discount dynamics will tell you whether the market treats RVI as a real investment vehicle or a meme trade on private market FOMO. If RVI trades persistently above NAV, Robinhood launches Fund II. If it doesn’t, the democratization thesis gets a lot harder to sell. Meanwhile, Fidelity and Schwab are watching closely. The $80 trillion retail asset pool that the World Economic Forum keeps citing is about to get a lot more pitches. That said, the key question now isn’t whether private markets are open to retail. It’s whether retail investors understand what they’re buying when the asset can’t be priced in real time and can’t be sold in a crisis. That’s the key here.

ICYMI: Robinhood: the $4.5 billion revenue dark horse that Wall Street still underestimates 🤷‍♂️📉 [deep dive into their Q4 & full year 2025 financials, what stands out, and why I’m still really bullish]

🧠 What else I’m watching

• Piraeus Launches AI Innovation Hub 🤖 Piraeus Bank has inaugurated an AI hub with Accenture and Anthropic to transition from isolated AI projects to an enterprise-wide capability. The hub will drive AI innovation across operations, customer experience, risk, and compliance, while modernizing the bank’s technology stack. Leveraging Udacity, Accenture’s AI training platform, Piraeus aims to embed AI skills organization-wide. ICYMI:

• Moonfare Launches AI Investment Strategy 🚀 Moonfare has unveiled an AI-focused early and growth stage strategy, offering diversified exposure to artificial intelligence and transformative innovation. The portfolio combines 8–15 top-tier venture and growth managers - selected for their track records in scaling category-defining tech - and 8–15 co-investments sourced through Moonfare’s GP network. The strategy targets key themes like AI, cybersecurity, next-gen healthcare, deep tech, fintech infrastructure, and physical AI. ICYMI:

• Aurionpro Revolutionizes Trade Finance 🤖 Aurionpro has launched Fintra, an AI-native trade finance platform that automates document processing, compliance, clause recommendations, and risk scoring using specialized AI agents, while ensuring human governance at every decision. Built on Aurion AI -an AI-native banking software stack - the platform integrates with core systems like SWIFT and General Ledger. Fintra introduces the Confidence-Gated Handoff Protocol (CGHP), which evaluates each decision’s confidence, materiality, and regulatory compliance before routing to human oversight, with full auditability. The platform is now available for pilot engagements with banks in India, the Middle East, and Southeast Asia, with additional AI-native solutions planned for corporate lending, retail lending, and supply chain finance. ICYMI:

💸 Following the Money

• Polymarket is talking to investors to raise $400M in new funding at a $15B valuation, The Information reported.

• Seapoint, a financial admin platform for startups founded by Stripe’s former European CIO, has raised €7.5M in Seed funding, bringing total funding to €10M in just over a year.

• Latin American digital bank Plata has closed on a $405M Series C at a $5B valuation.

👋 That’s it for today! Thank you for reading, and have a relaxing Sunday! And if you enjoyed this newsletter, invite your friends and colleagues to sign up:

Subscribe now

Subscribe now

On April 17, 2026, Anthropic launched Claude Design - an AI product that turns a plain-English prompt into a polished interactive prototype, pitch deck, or full design system, powered by Claude Opus 4.7. Four days later, on April 21, OpenAI dropped ChatGPT Images 2.0 (gpt-image-2) - the first AI image model with native reasoning capabilities, which claimed the #1 spot on the Image Arena leaderboard by a margin of +242 points within 12 hours of release. That’s the largest lead ever recorded 😳

In the span of a single week, the entire early-stage design workflow was made obsolete.

For FinTech and AI founders, it marks a structural shift. Design agencies that once charged $20,000-$50,000 and six weeks for a brand identity and prototype now compete with a workflow that costs less than $10 and runs in an afternoon. Figma’s stock fell 7.28% to $18.84 on the day Claude Design launched for a reason - down over 80% from its post-IPO high of $142.92. The market understood immediately what the tools meant.

But here is the problem: neither tool alone is enough.

ChatGPT Images 2.0 is the world’s most capable creative explorer and raw visual asset generator - but it has no design system tooling, no component library output, no prototype builder, and no code handoff. Claude Design is the world’s most capable systematic design builder - but it has a strict weekly usage limit that most users exhaust in a single complex session before getting anywhere near a final output.

When you understand the gap between what each tool does well, the answer becomes obvious: use them in sequence, not interchangeably. The seven-stage workflow in this guide is built exactly around that insight. It is the system that lets you enter Claude Design with a finalized, fully specified creative direction - and one-shot a production-grade design system and prototype without wasting your weekly quota on exploratory iteration.

Here’s what this deep dive covers:

• A complete technical breakdown of ChatGPT Images 2.0 (gpt-image-2) - what actually changed from prior models, the new reasoning/thinking mode, the 8-image batch generation capability, the near-perfect text rendering, and what it means for FinTech UI work

• A complete breakdown of Claude Design - capabilities, usage limits, export formats, the Claude Code handoff pipeline, and the Figma/Canva integration picture

• The 7-stage workflow that sequences both tools for maximum output quality and minimum quota burn - from strategic brief to deployed product

• FinTech-specific design principles: trust architecture, regulatory signal placement, typography as credibility infrastructure, and how to avoid the Stripe monoculture that makes most FinTech products invisible

• A complete, ready-to-use prompt library - 8 prompts tuned specifically for FinTech and AI product companies, covering logo exploration, brand guide generation, design system specification, dashboard prototyping, and investor deck creation

• Cost management strategy across both tools, including API pricing tiers and Claude Design’s weekly quota mechanics

• Known limitations and workarounds - covering style drift, prompt auto-rewrite, Figma round-trip, Canva flat-layer export, and the Opus 4.7 token cost increase

• Advanced use cases: competitive positioning maps, KYC flow diagrams, investor pitch decks, and brand extension/merch

• The Claude Code handoff pipeline - how to go from Claude Design prototype to running production code in a single clipboard paste

Every founder reading this has had the same experience: the product is real, the traction is real, the team is real - but the pitch deck looks like it was made in PowerPoint at 2 am, and the product UI looks like a shadcn template nobody touched.

That gap costs deals. It costs user trust. It costs regulatory credibility.

This guide closes that gap.

Part 1: Understanding the Tools

ChatGPT Images 2.0 (gpt-image-2)

Read more

Good morning & happy Thursday! Today’s issue is especially exciting as we’re looking into AgentList, which just launched USVC, a retail VC fund stuffed with the most talked-about AI companies on earth (why it’s not that exciting as it seems, what’s really under the hood here & what to watch for next + bonus list of recently funded AI startups, their pitch decks and what they tell us about the market & top 10 AI startups to watch in 2026 & their decks inside), eToro that stopped being a broker and became a platform (what The eToro App Store is all about & why it’s a smart play + bonus dive into eToro’s financials & the ultimate guide to Claude Managed Agents inside), and Robinhood that just invested into OpenAI (why this Robinhood bet is really a bet on itself & what to expect next + bonus deep dive into Robinhood’s latest financials & how private market investing is changing). So let’s jump straight into the finnovative stuff 🌶️

AngelList’s retail VC fund charges 2.5% to let you buy what your Uber driver already knows about 🥲📉

Following the money 💸 USVC is what happens when venture capital’s liquidity problem meets retail investors’ FOMO. Naval Ravikant’s name is on the letterhead, the portfolio is stuffed with the most talked-about AI companies on earth, and anyone with $500 can buy in.

Exciting? For sure.

But the prospectus tells a story the marketing page would rather you didn’t read. Layered fees, subjective valuations, a portfolio manager compensated for asset gathering rather than returns, and liquidity terms that would make a non-traded REIT blush. We dug through all 61 pages so you don’t have to.

Let’s see what USVC is really all about, what it tells us, and what’s next.

Read more

Subscribe now

On April 8, 2026, we witnessed one of the most important AI releases this year so far - Anthropic launched Claude Managed Agents, a set of composable APIs that let you build and deploy production AI agents on Anthropic’s cloud infrastructure without building any of the runtime yourself.

That matters because building a production AI agent has never been a model problem. It has been an infrastructure problem. Secure sandboxing, session persistence, credential isolation, error recovery, and observability - engineering teams routinely dedicated 4–8 senior engineers for 3–6 months before a single agent reached production.

Managed Agents eliminates that entire layer. You define what the agent does. Anthropic handles everything else.

It’s that simple now.

Five enterprise customers were already in production at launch: Notion, Rakuten, Asana, Sentry, and Atlassian. Rakuten cut critical errors by 97% and accelerated major releases from quarterly to biweekly. Sentry went from root cause analysis to a merged pull request in weeks instead of months. Asana’s CTO said Managed Agents let them ship advanced AI Teammates features “dramatically faster” than any prior approach.

The platform charges standard Claude API token rates plus $0.08 per session-hour of active runtime. No servers. No containers. No DevOps.

Below is the most thorough breakdown of Claude Managed Agents available today:

• The internal architecture (including the “stale harness” problem that drove the design)

• A step-by-step build walkthrough with production-ready code

• All five multi-agent coordination patterns

• Detailed case studies from each launch partner

• Seven production-tested system prompt templates

• A full pricing analysis with cost scenarios at different scales

• A head-to-head comparison of every alternative (Agent SDK, Messages API, LangChain, CrewAI, AutoGen, and two open-source options)

• Use case playbooks for founders, operators, agency builders, and investors

• The current limitations you need to plan around, and

• What the research preview features tell us about Anthropic’s roadmap.

Let’s dive in.

Part 1: What Claude Managed Agents Actually Is

The Problem It Solves

Read more

Good morning & happy Tuesday! Today we’re looking into Coinbase and their App Store for AI Agents (why it’s really a float play for USDC & Base dressed up as a protocol, why it matters for Coinbase & what’s next + bonus deep dive into Coinbase’s latest financials & a full guide on building your first AI Agent From Scratch), and ARK Invest, which just deepened their patnership with Kalshi (what this move is truly all about, what does it tell us about the future of prediction markets + bonus dive into Kalshi inside). So let’s jump straight into the interesting stuff 🌶️

Coinbase’s App Store for AI Agents is a float play dressed as a protocol 👀📈

The news 🗞️ Yesterday, Coinbase-incubated x402 Foundation launched Agentic.market, a directory where AI agents discover and pay for services without logins, API keys, or subscriptions.

The protocol revives HTTP’s long-dead 402 “Payment Required” status code to settle USDC micropayments on Base, typically in the $0.001 to $0.04 range.

Early numbers: $49M cumulative volume, 1.8M transactions in the last 30 days, 480,000 agents transacting. Indexed providers include Bloomberg, CoinGecko, LinkedIn, AWS Lambda, and Alchemy.

The obvious read here is that Coinbase built a payments rail for agents. The sharper read is that this reprices SaaS unit economics, quietly routes stablecoin float to Coinbase, and structurally excludes Visa and Stripe from machine traffic before they realize the fight has started. Here’s the part most people will miss.

Read more

Subscribe now

Three days before Anthropic shipped Claude Design, Mike Krieger resigned from Figma’s board. The SEC filing was boilerplate: no disagreement, cordial departure. Figma’s stock even went up 5%.

That was the window.

When Claude Design went live on April 17, Figma fell roughly 7%, and Adobe fell in sympathy. Figma at its post-IPO peak was worth more than $60 billion. Today, it’s worth less than $10 billion. Brutal.

On top of that, the iShares software ETF IGV is now down 18% year-to-date on a trade traders now call the SaaSpocalypse. In simple terms, the market is now publicly repricing the value of sitting between an AI lab and shipped code.

You, the founder or operator reading this, don’t need to price the collapse. You need to use the tool.

Because for the first time in 15 years, a founder with no design background can produce a brand system, a pitch deck, a landing page, a mobile app prototype, and email templates in a single afternoon, all on-brand, all exportable to Canva or straight into production code.

Email marketing template done with Claude Design. Credit: @Salmaaboukarr via X

Thanks to Claude Design, Mercury got 90% of its component library inferred from its codebase in just 10 minutes. Brilliant went from 20+ prompts to two. Datadog compressed a week-long design cycle into one conversation. A solo operator even built a complete mobile app flow for $2 in extra tokens.

Personal dashboard made by Claude Design. Credit: @jerrod_lew via X

Importantly, what matters here is that the output quality is not a function of your design skill. It’s a function of how you set the tool up and how you prompt it. Most people will get this wrong and ship generic AI slop. A few will ship work that looks like they hired an agency.

This is the complete founder’s playbook. Here’s what’s inside:

• It walks through the setup that separates on-brand output from generic slop (including which of three ingestion paths to use & why most people pick the wrong one).

• Claude Design shortcut for skipping 20 hours of brand strategy with a free repository of 68 ready-made design briefs.

• Eight workflows with the exact prompts that work (pitch decks, landing pages, mobile app prototypes, brand systems inferred from your codebase, logo and visual identity, internal tools, email templates and marketing collateral, and frontier interactive prototypes with 3D and motion)

• A complete seven-step logo design flow tuned for Claude Opus 4.7, including the 32-pixel social avatar scale test.

• A separate set of master prompts for rigorous design work (discovery interview, staged design tokens, brand strategy before visual identity, production-grade UI screens, honest design critique, accessibility review with stated limits, and design-to-code handoff).

• The pro tips that separate 30-minute shippers from people who burn their weekly quota with nothing to show, and the specific gotchas that waste tokens.

• The market implications of what just happened to Figma.

The window to compound this advantage is roughly 6 to 12 months. By Q4, every competitor will have something similar. What matters now therefore is how quickly you learn to use this specific tool.

Founders, entrepreneurs & operators who set up their design system on day one, use the handoff bundle to compress design-to-production cycles, and publish brand systems to their organizations will compound through a dozen launches before parity arrives. The rest will catch up when they catch up.

Here’s how to be in the first group.

The big picture, compressed

Read more

If you’re not a subscriber, here’s what you missed this week:

1. The Definitive Guide to Perplexity Computer 🖥️ [How Perplexity’s 19-model AI Agent works, what it costs, and whether it’s worth $200/month in 2026]

2. How to Build an AI Agent from Scratch (With Working Code) 🤖 [The design framework, working code, and hard-won lessons that take you from “I want to build an AI Agent” to a working one - in under 60 minutes]

3. Skill Graphs: The Architecture That Solves the AI Agent Context Window Problem 🤖 [Your AI agent gets dumber the more you teach it. Skill graphs fix that - here’s the architecture, the cognitive science, and a complete build-from-scratch tutorial]

4. Inside Revolut’s PRAGMA: The Foundation Model Trained on 40 Billion Banking Events 🧠 [Architecture, performance benchmarks vs. Stripe, Mastercard, and Visa, regulatory risks, and why PRAGMA may be the most consequential AI bet in consumer finance]

5. OpenAI is building a personal CFO in plain sight 🤖📊 [what their Hiro Finance M&A is all about, why they needed them & how it stacks into their bigger long-term strategy + bonus deep dive into OpenAI’s Agentic Singularity strategy inside]

6. 10 AI Prompt Skills That Actually Change How ChatGPT, Claude, and Gemini Respond 🤖 [Stop telling AI to “write clearly.” Give it real frameworks with measurable constraints and watch the output transform]

7. The File That Turns Claude Code Into Your Best Engineer 🧠 [A complete guide to CLAUDE.md, rules, skills, subagents, and hooks - everything you need to make Claude Code your 10X Engineer]

8. One unreleased AI model just triggered a global financial emergency 😳🚨 [why Claude Mythos prompted the US Treasury, the Fed, the Bank of Canada, and UK financial regulators to hold emergency meetings with their largest banks, where the biggest risks are & what’s next + bonus deep dive into Anthropic’s leaked Claude source code & how I turned Claude Code into my 10X engineer]

9. American Express skipped the Agent Protocol wars and bought the risk instead 🤖🛡️ [what The Amex Agentic Commerce Experiences (ACE) is all about, why it matters & what to expect next + bonus dives into Shopify’s Agents, AI Agents now opening bank accounts & how to build your first AI agent from scratch inside]

10. Shopify just made every AI agent a Shopify Agent 🛍️🤖 [what it’s all about & why it’s a brilliant play + bonus deep dive into Shopify’s financials & a full guide of building an AI Agent from scratch]

11. Turn Claude Cowork Into Your Personal COO 🧠 [Most people are using the most powerful AI agent ever released to answer questions. Here’s the framework that turns Claude Cowork into an operator that runs your work while you sleep]

12. The One-Person Unicorn 🦄 [I built an AI operating system to run a startup with Claude]

As for today, here are the 3 fascinating FinTech stories that are changing the world of financial technology as we know it. This was yet another intense week in the financial technology space, so make sure to check all the above stories.

Anthropic is now banking infra 🤖🏦

The news 🗞️ Anthropic’s UK head Pip White told Bloomberg this week that Mythos, the company’s cyber-capable frontier model, will reach UK financial institutions within days through an expansion of Project Glasswing.

But the real story isn’t the rollout. It’s that cybersecurity parity for global banks now depends on who Anthropic decides to let in.

Let’s unpack this.

More on this 👉 Mythos has already surfaced thousands of zero-day vulnerabilities across every major operating system and browser in internal testing, some sitting undiscovered for over two decades. The UK’s AI Security Institute confirmed Wednesday the model is “a step up over previous frontier models” at simulating multistep cyber attacks.

Bank of England Governor Andrew Bailey thus called for rapid global regulator evaluation. Treasury Secretary Bessent and Fed Chair Powell had already pulled Wall Street CEOs into emergency meetings after the initial US launch partners (JPMorgan, Microsoft, Apple, AWS, Cisco) got access.

Zoom out 🔎 Anthropic is framing this as safety-first gating. That’s half true. Mythos is powerful enough that unrestricted release would be reckless. But the same restraint creates one tier of institutions with AI-hardened perimeters and another without. Banks inside Glasswing can now run continuous automated pentests against their own COBOL-and-API spaghetti before attackers do. Banks outside it are defending yesterday’s stack against tomorrow’s threats 🤷‍♂️

The uncomfortable part for anyone holding bank equity: this is not the only frontier lab with this capability, and it will not stay contained. Distillation, leaks, and state-actor equivalents are a question of when. The window where Glasswing partners hold asymmetric defensive advantage is measured in months, not years.

THE TAKEAWAY ✈️

What’s next? 🤔 Looking ahead, watch three things. First and foremost, whether the FCA and PRA formalize AI-assisted vulnerability testing into operational resilience rules, which would convert access into compliance. Second, whether cyber insurance pricing bifurcates along Glasswing lines; underwriters love observable moats. Last but not least, whether tier-two UK banks and building societies get access, or get left absorbing systemic risk on behalf of the majors. Anthropic’s London office scaling from 200 to 800 staff by Q1 2027 suggests they expect demand to outrun their ability to supply it. But the firms that wait to be invited are already behind.

ICYMI: One unreleased AI model just triggered a global financial emergency 😳🚨 [why Claude Mythos prompted the US Treasury, the Fed, the Bank of Canada, and UK financial regulators to hold emergency meetings with their largest banks, where the biggest risks are & what’s next + bonus deep dive into Anthropic’s leaked Claude source code & how I turned Claude Code into my 10X engineer]

Revolut GlobalHire isn’t an HR product. It’s a payments product 👥💵

The news 🗞️ Every Employer of Record (EOR) company has the same dirty secret: the hardest part of hiring someone in Manila or Tallinn isn’t the legal paperwork. It’s moving money across borders without losing 3–5% to FX markups and third-party payment rails. Revolut just made that the entire point.

Let’s unpack this.

More on this 👉 Yesterday, Revolut launched GlobalHire for its UK Business customers, offering full EOR coverage across 160+ countries at £499 per employee per month, undercutting Deel’s $599 industry standard.

The pricing gap is real but secondary.

The structural move here is that payroll runs natively through the customer’s Revolut Business account at interbank FX rates, zeroing out the currency conversion fees that every other EOR provider either charges or absorbs. Revolut processes $365 billion in annual transaction volume across 39 countries. The hardest operational layer in EOR is literally its core business.

Zoom out 🔎 The timing here is deliberate. GlobalHire sits atop two existing layers: Revolut Business (800K customers, 16% of group revenue) and Revolut People, the HRIS platform Revolut dogfooded internally for its own 10,000+ employees before commercializing it.

Together, they form something none of the pure-play EOR providers can replicate: a single system where a company’s financial infrastructure and people infrastructure share the same data backbone.

And that backbone now includes PRAGMA, Revolut’s foundation model trained on 24 billion banking events from 26 million users. PRAGMA’s product recommendation accuracy improved 40.5% over prior baselines, and its ability to detect recurring transactions means Revolut can identify existing international payroll flows inside Business accounts and target them for conversion before any sales rep picks up the phone. Win-win.

THE TAKEAWAY ✈️

What’s next? 🤔 Obviously, EOR giants like Deel ($1.4B ARR, $17.3B valuation) or Rippling won’t be displaced overnight. It has 35,000+ customers and is building owned entities in 100+ countries. But the mid-tier EOR market faces an existential squeeze. Omnipresent has already sold to Deel for ~$15 million, well below its total funding. Revolut’s entry with deep pockets and a captive distribution base of 800K businesses will only accelerate that consolidation. But the second-order effect worth tracking is far more interesting here. Every GlobalHire payroll event feeds new token types into PRAGMA, enriching the model’s understanding of each business customer, which thus improves credit scoring, cross-sell targeting, and fraud detection across the entire Revolut stack. In other words, the B2B Super App thesis just got its strongest proof point. Therefore, Revolut isn’t entering EOR per se. It’s making employment a financial product, and the companies that built great HR software on top of bad financial plumbing are about to find out what that means for their margins. Yet another reason to be bullish on Revolut.

ICYMI:

Revolut is the most profitable FinTech on Earth, and it hasn’t even started lending yet 🤯📈 [unpacking the fee-income fortress that makes Revolut structurally different from every competitor, the lending optionality coiled inside that balance sheet, the regulatory dominoes now falling in sequence, and whether $75B is expensive or whether the market is about to find out it was way too cheap + more bonus reads on Revolut, how it’s leveraging AI & bonus deep dives into the latest financials of Nubank, SoFi, Robinhood, and Coinbase inside]

AI Agents started opening bank accounts now 🤖🏦

The news 🗞️ For two years, AI agents have been able to write code, browse the web, send emails, and reason through complex problems. The one thing they couldn’t do was spend money. Meow just fixed that.

Let’s unpack this.

More on this 👉 Last week, fintech startup Meow (founded 2021, ~$30M raised from Tiger Global, QED, and Lux, already managing billions in assets) launched a native MCP server at meow.com/mcp that turns its entire banking stack into a promptable tool.

Any MCP-compatible agent - Claude, ChatGPT, Gemini, Cursor - can now open FDIC-insured business checking accounts, issue virtual and physical corporate cards, execute ACH, wire, and crypto transfers across 50+ currencies, pull balances, audit spend, and manage invoicing.

No dashboard. No login. One natural-language prompt.

Zoom out 🔎 The weekend adoption numbers tell the story brilliantly: CEO Brandon Arvanaghi reported agents “signing up in droves” within 72 hours of launch.

Turns out, the demand was already there, just waiting for infrastructure to catch up.

But what makes Meow’s approach credible rather than reckless is the guardrail architecture. Agents operate under scoped permissions with role-based controls, daily limits, and mandatory human approvals (via chat, SMS, or Telegram) for high-value actions. Account numbers stay hidden from the LLM. Every transaction is auditable. Meow isn’t a bank itself; Cross River Bank and Grasshopper Bank provide the FDIC backing, which means the compliance rails were already built before agents ever touched them.

More importantly, the real unlock here isn’t convenience. It’s that finance was the binding constraint on end-to-end agentic workflows. An agent that can research a market, build a product, and write marketing copy, but can’t pay a contractor or invoice a client is still a toy.

Now it isn’t.

THE TAKEAWAY ✈️

What’s next? 🤔 Looking ahead, watch for three things over the next two quarters. First, a fast-follow race: Mercury, Brex, Ramp, and Stripe will all ship MCP endpoints or lose developer mindshare to a company a fraction of their size. Second, the first “agent gone wrong” wire transfer story, which will test whether Meow’s guardrails hold and shape how regulators frame liability for autonomous financial actors. Third, and most interesting: multi-agent payment flows where one agent hires and pays another. That’s where agent-native banking stops being a developer convenience and starts looking like infrastructure for a genuinely new type of economy. The companies building for that world today won’t have to retrofit later. Agentic is the future.

ICYMI:

What else I’m watching

• Lloyds Installs AI in Boardroom 🤖 Lloyds Banking Group has introduced an AI “board bot,” developed by Board Intelligence, to assist its board in reviewing confidential materials and preparing for meetings, aiming to reduce human bias in decision-making. The bot provides guidance across cybersecurity, sustainability, financial analysis, and M&A. ICYMI:

• Nu Mexico Reaches 15M Customers 🇲🇽 Nu NU -2.60%↓ Mexico has surpassed 15M customers, becoming one of Mexico’s top three financial institutions by user base and achieving 36% year-over-year growth - even faster than its Brazilian operations. Nu’s nationwide success stems from its 24/7 digital, branchless model, resolving over 30,000 needs daily while maintaining human connection. The Flywheel model drives efficiency: scale and technology refine products, reduce costs, and reinvest in user experience. Financial performance reflects this growth, with a 61% rise in the credit portfolio and 21% growth in deposits. ICYMI: Nubank’s Q4 2025: Latin America’s digital banking juggernaut is just getting started (but the market already knows it) 😤📈 [deep dive into their Q4 & full 2025 year financials, breaking down the most important facts and figures, understanding what they mean & what’s next for ]

• GoCardless Hits Profitability Milestone 📈 GoCardless has achieved its first profitable quarter, driven by a 22% revenue increase and 27% reduction in net losses for FY25. Cost controls, including layoffs and relocating roles to Eastern Europe, contributed to a 55% decrease in losses the prior year. The firm posted its first EBITDA-positive quarter in Q4 FY25 and expects its first profitable year in FY26. Losses before tax narrowed to £24.2M from £31.2M, despite restructuring costs. Acquired by Mollie in December, GoCardless processes over $130B annually for 100,000+ businesses across 30+ countries. ICYMI: Mollie seals GoCardless takeover, creating European payments powerhouse to rival Stripe and Adyen 💳🇪🇺 [what it’s all about & why it makes a ton of sense + bonus dive into Stripe’s recent M&A game, and how they want to own the full AI payments stack]

💸 Following the Money

• Polygon Labs is in talks to raise up to $100M for a new stablecoin payments business, according to a report from The Information.

• Deutsche Börse has spent $200M to buy a 1.5% fully diluted stake in crypto platform Kraken, deepening an existing strategic partnership between the two firms.

• Ralio, a UK startup building a payments platform purpose-built for AI agents, has raised $2.5M in an oversubscribed pre-seed funding round.

👋 That’s it for today! Thank you for reading, and have a relaxing Sunday! And if you enjoyed this newsletter, invite your friends and colleagues to sign up:

Subscribe now

Good morning & happy Friday! Today we’re looking into American Express that did something interesting in the Agentic AI space - it agreed to pay when the agents mess up (what The Amex Agentic Commerce Experiences (ACE) is all about, why it matters & what to expect next + bonus dives into Shopify’s Agents, AI Agents now opening bank accounts & how to build your first AI agent from scratch inside), and Anthropic, which now is basically a banking infrastructure (the real story behind UK financial institutions getting access to Claude Mythos & what to expect next + bonus dive into how one unreleased AI model just triggered a global financial emergency). So let’s jump straight into the good stuff 🌶️

American Express skipped the Agent Protocol wars and bought the risk instead 🤖🛡️

Following the trends 🤖 While Visa and Mastercard spent the last quarter publishing specs for AI agent authentication, American Express did something more interesting: it agreed to pay when the agents mess up.

Let’s take a closer look at this, understand what ACE is all about and why it could be far more important than anything else.

Read more

Good morning & happy Thursday! Today’s issue is incredibly exciting as we’re looking into OpenAI, which is now building a personal AI CFO (what their Hiro Finance M&A is all about, why they needed them & how it stacks into their bigger long-term strategy + bonus deep dive into OpenAI’s Agentic Singularity strategy inside), Revolut’s GlobalHire launch that’s actually a payments product disguised as an HR product (understanding the real value behind it, how it perfectly fits their Super App play & what it means for Deel, Rippling and other EOR platorms + bonus deep dive into Revolut’s PRAGMA AI model & their 2025 Financials inside), and Shopify, which just made every AI agent a Shopify Agent (what it’s all about & why it’s a brilliant play + bonus deep dive into Shopify’s financials & a full guide of building an AI Agent from scratch). Let’s jump straight into the awesome stuff 🌶️

OpenAI is building a personal CFO in plain sight 🤖📊

The news 🗞️ AI giant OpenAI just acquired Hiro Finance this week, its second personal finance startup in just six months. The pattern is the story here, and the pattern is much bigger than finance.

Let’s take a closer look at this M&A, understand how & why it matters for OpenAI’s long-term strategy, and what to expect next.

Read more

Subscribe now

Shortly after announcing their phenomenal 2025 financial results, Revolut has published PRAGMA (arXiv: 2604.08649), a family of encoder-style Transformer foundation models pre-trained on banking event sequences from roughly 25 million users across 111 countries. The training corpus spans approximately 40 billion events and 207 billion tokens.

What’s interesting here is that the architecture replaces the old paradigm of siloed, task-specific models with hand-crafted features: a single shared backbone transfers learned representations across credit scoring, fraud detection, lifetime value prediction, communication engagement, and more.

The results are quite striking - a +130.2% lift in Credit Scoring PR-AUC, +64.7% improvement in Fraud Recall, and +79.4% gain in Communication Engagement PR-AUC over production baselines.

But the performance numbers are not the story here.

PRAGMA is a structural bet on a specific theory of value creation: that rich, longitudinal behavioral data from a Super App generates embeddings universally better than anything hand-crafted per task, and that the competitive moat in AI is no longer algorithms but proprietary event scale.

Within the past 12 months, Stripe, Mastercard , Visa , and now Revolut have all published or announced foundation models for financial data. That convergence is definitely not a coincidence. It marks the beginning of a new competitive layer in financial services - one where the quality of behavioral representation matters more than the sophistication of any individual model.

PRAGMA is arguably the most ambitious of the group: it is the only model that fuses multiple event sources (transactions, app navigation, trading, push notification interactions) into a single user-level embedding, and the only one built by a consumer neobank rather than a payments network.

For anyone building, investing in, or competing with modern financial infrastructure, what matters now is understanding:

• How PRAGMA’s three-encoder architecture actually works, and where it breaks down

• Why it delivers a 130% improvement in credit scoring but a 47% degradation in anti-money laundering (AML)

• How it compares, on precise technical terms, to Stripe’s PFM, Mastercard’s LTM, and Visa’s TransactionGPT

• Which regulatory constraints could keep PRAGMA’s best results out of Revolut’s highest-value markets for years

• Whether traditional banks can replicate any of this, or whether their organizational structure makes it structurally impossible

• What the minimum data scale threshold is for fintechs and neobanks that want to build their own foundation models

• Where the commercial opportunities are for startups - from thin-file credit scoring to explainability tooling

• How the foundation model arms race between Stripe, Mastercard, Visa, and Revolut reshapes the competitive landscape for the entire industry

This deep dive works through each of those in detail.

The Problem PRAGMA Was Built to Solve

Read more

Good morning & happy Tuesday! Today’s issue is super exciting because we’re diving into how one unreleased AI model has triggered a global financial emergency (why Claude Mythos prompted the US Treasury, the Fed, the Bank of Canada, and UK financial regulators to hold emergency meetings with their largest banks, where the biggest risks are & what’s next + bonus deep dive into Anthropic’s leaked Claude source code & how I turned Claude Code into my 10X engineer), and AI Agents that now started opening bank accounts (how banking FinTech Meow turned its entire banking stack into a promptable tool for Claude, ChatGPT & Gemini, why it’s interesting & what it tells us about the future + bonus deep dives into AI moves & strategies from Visa, Ramp, Stripe, and a full guide on how to build your first AI Agent from scracth inside). So let’s just jump straight into the interesting stuff 🌶️

One unreleased AI model just triggered a global financial emergency 😳🚨

The BIG News 🗞️ Within 72 hours of Anthropic’s latest AI release, the US Treasury, the Fed, the Bank of Canada, and UK financial regulators all convened emergency meetings with their largest banks.

Not about rates. Not about liquidity. About a single AI model 🤖

On April 8, Treasury Secretary Scott Bessent and Fed Chair Jerome Powell pulled the CEOs of Citi , Goldman , Morgan Stanley , Bank of America , and Wells Fargo into a closed-door session at Treasury headquarters. By Friday, the Bank of Canada had assembled its own banks and regulators. The Bank of England, the FCA, and HM Treasury are now coordinating with the National Cyber Security Centre, with a meeting of leading British banks, insurers, and exchanges expected within two weeks.

Three countries, three separate convenings, one cause: Anthropic’s Claude Mythos Preview, announced April 7, can reportedly autonomously discover and exploit zero-day vulnerabilities across every major operating system and web browser.

Let’s take a quick look at this, understand why it matters, and what to expect next.

Read more