6x1? Here it's 24/7
The whole story: from the Saturday night the idea came up to the first nights the factory worked on its own.
published · October 3, 2026Read the chapter →Personal project · built in public
Relax, I'm not the one working around the clock. It's an agentic software factory I put together as a hobby: one agent writes the code, another tests it, another reviews it, and nothing reaches main without a rule of mine or my approval. While I sleep, it keeps going.
The story
"What if I could open several AI sessions at the same time?" Within a few days, the question became a small agentic software factory. And then the most curious part happened: the tool started developing itself.
An app that orchestrates Claude Code, Codex and Grok in a real development flow, with cross-review, tests and a gate where only I decide.
One AI implements, another from a different family reviews, a third writes the tests. And a supervisor that only steps in when something fails.
I wrote down the backlog, and the issues started being implemented by the tool itself. From then on, I followed along instead of writing code.
An AI Product Owner laid out the plan in stages, and the roadmap ran until morning. I went to sleep and woke up to 10 tasks delivered, tested and merged.
By day I say the ideas, by night they become code. A night watch unblocks what it can, within what I allowed, and leaves the decisions to me.
A day in the factory
The roadmap runs stage by stage: implement, review, test, merge. What breaks, the watch tries to unblock; what needs me, waits.
I review what got done by its result: screenshots, evidence, cost. I approve, ask for changes or discard.
I talk to the Product Owner, who reads the code before giving an opinion, and accept the plan for the next 24 hours.
New ideas come in by text or voice, get triaged and join the queue without waiting for the next sprint.
Recordings of the app in simulation mode.
The chapters
Nine chapters, each with its data and its mistakes. Read the ones already out and subscribe to get the next ones.
The whole story: from the Saturday night the idea came up to the first nights the factory worked on its own.
published · October 3, 2026Read the chapter →On day four, the tool started generating more agent work than I did. A few days later, over 90% came from it.
coming soonThe first full night: a roadmap in stages, not a single prompt from me, and 10 tasks delivered, tested and merged by morning.
coming soonAdding up the time of each step, testing took longer than development. The next big saving is in testing.
coming soonThe most expensive run delivered nothing: an AI polling a status that never changed. Waiting is a job for code.
coming soonTwo agents conflict just by adding a line at the end of the same list. With AIs in parallel, file layout becomes architecture.
coming soonWhen generating gets cheap, a working increment becomes the cheapest specification there is.
coming soonAll the state in readable files: when two copies of the app scrambled the plan, the AI read it, understood it and fixed it in two minutes.
coming soonPiece by piece, the tool started rewriting its own code. Is it still the same software? And what is left for me?
coming soonNewsletter
One email when a new chapter is out: what worked, what broke and what the factory learned. No spam, unsubscribe anytime.
The next chapter will reach your inbox.
AI Arena, Write & Review, workflows, Product Owner, night watch, analytics: every module with a video (page in Portuguese).
See every feature →