Factor Capital Update - April 2026

This month felt like the loudest chorus in my universe, centered on the growing adoption of agents, specifically OpenClaw and its peers.

I am writing this on my way back from a trip and walking through the airport terminal, something struck me. On the laptop screens around me in the terminal and in flight, I saw a woman tabbing between QuickBooks and a spreadsheet. A guy building a slide deck of quarterly results for a corporate presentation. Someone copying status updates from Gmail into a project management tool. I felt like I was looking into the past.

We are not quite in the "remember when we had to do that manually" phase yet. But we are close. Most of the people clicking through those workflows have no idea.

This is not optional

At most companies, having an assistant is a luxury afforded to the C-suite, and productivity tools like Calendly aim to make some version of this luxury more widely available. I have been using OpenClaw for the past month. It is hands down the best chief of staff I have ever worked with. It gives you a glimpse at the life of Miranda Priestly in The Devil Wears Prada, where every moment of her day is perfectly orchestrated down to the assistant whispering in her ear about the approaching guest at the gala.

And yet, these tools, while powerful, are still raw and early-adopter territory. But it is very clear that we're not far off from this being critical corporate infrastructure.

The fastest-growing tech companies have already made the shift. Stripe built Minions, autonomous coding agents that produce over 1,300 pull requests (bundles of new code ready to be added to a product) per week. All human-reviewed, zero human-written. The code they touch supports more than $1 trillion in annual volume. Ramp built Inspect, an internal agent now powering 30% of their merged pull requests. Nobody told engineers to use it. They just did, because it was faster.

The pattern extends outside of engineering too. Pedro Franceschi at Brex runs the company and his personal life with a team of OpenClaw agents covering recruiting, management, sales, and operations. USV deployed an entire roster of named agents across their investment team using a hosted tool called Tasklet: a round-the-clock deal analyst, an email logger, a meeting scribe, and a legal counsel agent. Fred Wilson said they are going "fully agentic." Their receptionist even uses agents to automate guest registration with their building’s front desk system.

There is zero excuse for any knowledge worker to still be doing data entry or manually updating systems of record. Regulators and compliance teams will still require a human to click "approve" for liability reasons on certain types of work. But the actual generation of that data will be agentic. The human becomes an editor and approver. Employees should be building products, speaking with customers, or directing agents. The companies that restructure around this will accelerate. The ones that wait will lose. Period.

Think of the shift from bricks and mortar to e-commerce. The web-native companies put Sears and Kmart out of business. By the time those companies tried to catch up, the race was over. Or think of quant funds. When Renaissance and, my former employer, Two Sigma showed up, they restructured equity markets with speed and data no individual manager could match. Many generalist, fundamental long-short managers got left behind. The same dynamic is playing out now. Build an AI-native org structure or watch your competitors do it first. Become an AI-native worker or find that your peers have raced past you.

The agent-first org chart

What does this restructuring look like in practice? Some teams become entirely agentic. A marketing org today can be one person orchestrating content, campaigns, and analytics through OpenClaw. As I wrote last month, Block cut over 4,000 people and the stock jumped 20%. Dorsey is explicitly re-architecting to be agent-first and discussed exactly how he sees Block becoming a flat organization with every employee reporting directly to the CEO through agentic leverage. Tools like Paperclip build upon OpenClaw and enable you to launch a whole company of agents with various roles, like a traditional company. I experimented with this to have it launch my pet project babysitter app. The results are pretty impressive.

On the engineering side, teams are building what they call "harnesses," secure corporate versions of these tools running in sandboxed environments where the AI cannot accidentally break production. Stripe's Minions evolved from an internal fork of Block's open-source Goose agent. Ramp gives agents the same access as human engineers: databases, automated testing and deployment pipelines, Slack, GitHub. This is how these companies operate now.

Even in non-technical, blue-collar work, these agents will play a role, providing a powerful interface for accessing replacement parts inventory, guided documentation around product manuals and repairs, and interfacing with routing and dispatch agents at the central office before too long. It feels hard to underwrite right now until you've seen how natural these systems are to interact with, even for the least technical user. You literally just text with your assistant. It is the setup that's the challenge today.

As a result, traditional functional silos are collapsing. We see this in our portfolio. One CEO told me he now has a product head leading his sales organization. The sales team did not disappear, but they do not need a traditional sales leader. And every team member is aware that they need to be adopting agentic tools in their daily work, or they'll be managed out. Every call on the team is recorded and feeds the system in real time, surfacing bugs, feature requests, and sales challenges. The business has live pipeline visibility without anyone doing manual entry. The team spends its time thinking about growth, and that thinking flows up to a product leader building tools for both the sales team and customers. When you build in bits rather than atoms, everything blurs with product.

The browser is getting in the way

When you use these tools seriously, you realize the browser is a blocker. This is a thread I have been pulling on since last year.

The web was built around session management, CAPTCHA, and traditional billing and onboarding funnels. When you delegate work to agents, those protections become walls. The woman in the airport using QuickBooks needs a graphical interface to see fields and click dropdowns. An agent just needs a structured command. Services need to expose programmatic interfaces that agents can talk to directly. The browser, with all its human-verification scaffolding, is solving a problem from the last decade.

When agents operate programmatically, they need a payment mechanism that works the same way: instant and gated by permissions.

Agents need wallets, not passwords

Ironically, at a moment when crypto sentiment feels like it is in the worst place I can remember, this might be its most natural use case. You give an agent a wallet and authorize spending through cryptographic signing. Same pattern as approving a transaction on a DeFi exchange, or approving a Claude Code command before it runs.

In February's update, I described spinning up a wallet for an agent funded with USDC, a crypto token pegged to the US dollar. The agent spends programmatically, but the amount cannot exceed the balance. That was the concept. Now it is real.

Last year, Stripe and Paradigm launched the Tempo blockchain. Its main network went live in March with Machine Payments Protocol (MPP), an agentic payments layer. You fund an agent's wallet to pay for services using micropayments. Money unlocks as work gets done. When agents do the work, payments flow at the task level, not billing cycles.

The platform race

The competitive picture is taking shape. OpenClaw is open-source and user-controlled and seems on track to be the Android or Linux of agentic operating systems. OpenAI acqui-hired OpenClaw's creator to build this orchestration into ChatGPT. Google is weaving agents through Gemini, and the release of their powerful open-source Gemma 4 model feels like a prelude to their future entrance here. Google has the distribution to make this a dominant platform if they execute, and lately they have been executing at a high level.

Anthropic is doing something different, and the numbers are hard to argue with. They shipped 74 releases in 52 days through February and March. They went from $1B to $30B in annual recurring revenue in under 18 months.

Cowork, their increasingly powerful and hostile answer to OpenClaw, was mostly built by Claude Code itself in ten days. The product builds the product. That kind of feedback loop compounds in ways that are genuinely hard to keep up with, especially when we now realize internally they've been using a model called Mythos, soon to be released to the public, that is a level better than current frontier models that defies belief. When models like Mythos get attached to agentic platforms, things will get very interesting and perhaps a bit scary.

What matters now is which product makes agents most useful across the broadest set of work. And companies will orient around customizing and orchestrating these agentic platforms to meet their specific needs and effectively 10x their workforce with zero new hires and exclusively added compute spend. This is an entirely new paradigm where labor and compute costs in an ordinary services business will begin to converge.

What this means

That airport terminal felt like a snapshot of a moment about to pass. Like Matt Shumer wrote in January with his pre-COVID analogy. Stripe ships 1,300 agent-written pull requests a week. USV has a roster of agents who run its investment operations. Anthropic shipped 74 releases in 52 days. The tools are here. The companies restructuring around them are pulling ahead. The rest are still building slide decks and Excel models that may or may not point to their coming obsolescence.

Thanks as always for reading.

-Jake

Jake Dwyer
Founder
Factor Capital