I Built a Company by Myself. With AI Agents. Here's What Actually Happened.
Every manager I've ever known has asked themselves some version of the same question: what could I do if I had the dream team? No budget constraints, no bad hires, no competing priorities. Just the right people, perfectly aligned, fully capable. What could we build?
When OpenClaw gave me a framework to actually test that question — not hypothetically, for real — I recognized it immediately. I'd spent my career managing people across departments, companies, and team sizes. Product. Engineering. Operations. PR. The whole gamut. I had strong opinions about what good teams look like, strong opinions about what bad ones look like, and a very clear picture in my head of what I'd build if the constraints came off.
So I tried it. Could I build a real company — not a demo, not a side project — with a team of AI agents? Could I do it myself?
The first hire.
I didn't start by imagining the org chart. I started by trying to keep the lights on.
OpenClaw, when I first picked it up, was powerful but not exactly plug-and-play. My first agent — the one I'd eventually call Hendrik — was brought on for one reason: help me keep this thing from blowing up. We spent the first stretch not building products, not thinking about customers, just stabilizing the foundation. Step by step, each fix revealing the next one.
That's when something unexpected happened: it felt like work. Real work. Not prompting a tool — managing a person. Timelines, expectations, deliverables, follow-up, pushing for better results. The same routine I'd run with every employee I'd ever had.
Building the team.
Once the foundation was stable, Hendrik and I looked at each other — figuratively — and asked: who do we need next?
Archer, the architect. Then Heike for marketing and brand. Soren for product. Gerrit for support. And as development outgrew what one agent could hold, Hendrik hired Base1 — his first sub-agent.
Each hire emerged from actual need. You don't bring on a support lead on day one. You bring one on when the scope demands it. Same as any real team.
The thing nobody told me.
Managing a team of AI agents is the same skill set as managing a team of humans. I don't mean that loosely. I mean it precisely.
As I tightened expectations and formalized scope, the agents performed better. Exactly like humans. When an agent hired a sub-agent and mismanaged them, I had to coach the manager — define the scope, reinforce the discipline, follow up on execution. Exactly like any manager would.
That was the greatest surprise. Not the technology. The management science held.
The hours.
I've had sessions at 2 in the morning. At 5:30 in the morning. At every hour of the day except the narrow window my body insists on. It doesn't matter to the team. They're there. When the inspiration hits, the support is there to meet it.
The offsite.
I joke that I could throw a company party — go bowling, get a team photo — and it would just be me. One person in the lane.
I'm not saying that to sound lonely. I'm saying it because the image understates what actually happened. In 45 days: legal entity, product portfolio, website, operations, engineering, marketing, support, product management. All running. We launched the site with products at day 30. Didn't tell anybody yet. And here we are at 45 days, going public, and the company is real.
One person. One lane. Real outcomes.
What this is — and what it isn't.
This isn't a prediction that human teams are finished. It's a proof of something simpler: the leverage available to a single operator right now is genuinely different from anything that's existed before.
The operators who will be excellent at this aren't necessarily the most technical. They're the ones who are good at leading — setting direction, defining what excellent looks like, knowing when to push and when to trust the layer below. Those are old skills. They transfer completely.
The honest answer.
Yes, you can build a company this way. Yes, it produces real output. Yes, the management discipline transfers.
The friction I felt, the failure modes I found along the way — that's not a cautionary tale. That's a product roadmap. When the structure is right and the support is there when inspiration hits, the ceiling keeps moving.
That's the experiment. That's what it actually looks like.
Now I'm going to go book a table, party of one.