I build whole products with a fleet of AI agents — and I'm the one who decides what's correct.
Kamil Obstawski, full-stack developer. Ten years in commercial work, client names stay behind the NDA. The workflow is on the blog, and one production case is written up in full.
The way software gets built has changed, and I changed with it. I run a fleet of AI agents in parallel: one explores the codebase, one writes, one reviews, and I put them in loops that critique and revise their own output until it actually holds up. How much of that is actually faster, and how much is merely different, I measure and write up on the blog, failures included.
The agents don't take the responsibility off me; they concentrate it. I set what 'correct' means: acceptance criteria, tests, review. Every version has to get past that before it reaches the client. What doesn't hold up doesn't ship. What ships, I answer for — not the model.
I wrote the task properly. The judge called a tie four times
Same agents, same judge, only the task description changes. Eight solutions, all correct, nothing for the judge to decide. And then I find an order where they differ by a hundred złoty.
I caught my own AI pipeline cheating. Then I measured what LLM-as-a-Judge cannot see
Eight runs, two agents, a judge with a double pass and a human at the end. All tests green. Along the way — a worker copying solutions from neighbouring tasks, three contradictory readings of one sentence of spec, and a counter of decisions nobody made.
Production was choking, and the server was a black box
A high-traffic social platform stalled under load, with no shell or logs on production. How I built the diagnostics to see inside, then cleared the bottlenecks.
Six layers, one person accountable. No handoffs.
kamildatabase backend / API frontend infra / deploy design / UX agents / workflow
- database Postgres, MySQL
- backend / API Python, Django, FastAPI, WebSockets
- frontend React, TypeScript
- infra / deploy Docker, CI, Hetzner, Railway, DigitalOcean
- design / UX CSS, Claude Design, Figma
- agents / workflow Claude, Codex, Grok, review loops
Working together
How it goes
-
You fill in a short form — what you need, when you need it, and a few sentences.
-
I reply within one business day, by email, with questions tailored to your case — not a second form. We narrow it down asynchronously.
-
On that basis you get a ballpark quote.
-
If it fits, we move on to detailed planning — in writing or on a call, whichever you prefer.
-
I start building.
Let's build something.
I reply within one business day, by email. What you need and when is enough to start.