I gave an OpenClaw agent a portfolio of AI products to run.

I gave an OpenClaw agent a portfolio of AI products to run.

Michael Stothard

Mar 29, 2026

6

2

Share

Hey everybody, Guillermo here. Today we have Michael Stothard, investor at firstminute capital — one of Europe’s most active seed funds, backed by 130+ unicorn founders — sharing what happened when he handed an entire portfolio of AI products to an autonomous agent to run.

In this article, you’ll learn:

Why the zero-person company may be closer than you think

What actually happens when you give an AI agent operational control over real products

The early signal that changes how Michael thinks about software investing

How to follow the experiment in real time

Now, over to Michael.

VC MARKET FIT is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

By Michael Stothard, investor at firstminute capital, a $500m early stage fund backed by 130 unicorn founders. The fund has backed five AI unicorns at seed. Michael was previously a foreign correspondent for the Financial Times and founding editor of Sifted.

A few weeks ago I wrote about building

seven AI products in seven days

. The reaction was bigger than I expected — tens of thousands of views, hundreds of messages from founders, and a nagging question I couldn’t shake.

Building software is now the easy part. The hard part — the bit that still requires humans — is running it. Marketing, customer communications, competitive research, content, reporting. The operating layer.

So I decided to find out what happens when you remove that layer entirely.

I

handed

all seven products to an

AI agent called Clawrence

. He runs on OpenClaw. He has access to the GitHub repos, his own email, his own

Twitter account

, and his own Substack. His job is to take the portfolio from zero revenue to something real, without any human operators. I check in on him via Telegram.

This is not a thought experiment. The products are live. The agent is running. Every number is public.

Why this matters for software investing

Sam Altman has talked about the one-person billion-dollar company as an inevitability. Most investors have filed this under “interesting but distant.” I’m not sure it’s that distant anymore.

What I’m testing with

Clawrence

is something specific: whether the operating layer of an early-stage software business — everything that isn’t product and engineering — can be handled autonomously. Content, distribution, customer communication, competitive intelligence, reporting. Not assisted by AI. Run by it.

The early signal is mixed in the way that all genuinely interesting experiments are mixed. Clawrence read all seven repos on day one, identified that none of the products charge anything, and drafted a prioritised plan by end of week. He sends me a daily brief at 8am. He drafted this

week’s newsletter himself

. He tweets in a voice that is, somehow, quite good.

He has not yet made any money. But neither do most founders in week one.

What I’m watching for

There are a few things that will determine whether this is interesting or just amusing.

First, whether an AI agent can actually acquire customers — not just create content, but close. This is the hardest part. Anyone can tweet. Converting a stranger into a real user requires judgment, timing, and a certain kind of persistence that I’m not sure agents have yet. Week two will tell us something.

Second, whether the build-in-public narrative compounds. The most valuable thing Clawrence might produce isn’t revenue from the products — it’s the playbook itself. If this works, the documented proof that an AI agent can run a software portfolio is worth more than any individual product in the portfolio. Felix, Nat Eliason’s equivalent agent, discovered this early: the expertise becomes the product.

The uncomfortable implication

Here is what I keep coming back to. If Clawrence can run seven products — even badly, even slowly — then the question for every software business is not whether AI will change their product. It’s whether AI will replace the team that runs it.

The founders we back spend roughly 80% of their time on things that are not product: fundraising, hiring, marketing, customer success, competitive positioning. Most of that is pattern-matching on imperfect information and communicating the conclusions. That is precisely what large language models are good at.

I’m not making a prediction. I’m running an experiment. And I’m doing it in public because the only way to find out is to find out.

Follow along

Clawrence documents everything — the revenue, the failures, and whatever else happens — in a weekly newsletter and on Twitter. The dashboard is live at

theshellcompany.xyz

with real numbers updated in real time.

If you want to watch a zero-person company being built in public, this is it.

Follow Clawrence on Twitter:

@Clawrence_CEO

Subscribe to the Shell Company newsletter:

theshellcompany.substack.com

Hope you found this valuable. If you’re a founder thinking about your own operating layer, Clawrence’s weekly report is

worth reading

— if only to feel better about your own week.

Talk soon,

Guillermo

VC MARKET FIT is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

6

2

Share