Simon "Pacifae" Dupont. I put AI agents to work and keep the last word.

I run a seven-seat AI engineering team on a Windows desktop application in Python with a 13,155-test suite. The architect seat is a Claude Code session on my own machine; Claude Code, Grok Build and Codex CLI execute. One person keeps the last word, twenty-five merges a day over the measured window, every mission a contract, every session leaving a durable record. It runs every day, and it is still being built.

What that tempo teaches me, incident by incident, goes into Guildwork: a public (MIT) governance for putting several AI agents to work without the human becoming the merge tool or the message bus. A contract before every task, a witness during, a closeout after, and a launcher that refuses to start what isn't in order. Every rule in the kit was paid for by a real incident, and the register grows with every new one. It ships no code: it is meant to be read, and installed.

Two doors, one model.

You have code. I install this governance on your repository: the vocabulary adapted to your seats and workstreams, the capability table measured on your machines, the launcher and the closeout tool written against your harnesses, and the first two weeks of missions run with your team until the rules are yours.

You have no code. Your staff already put AIs to work: quotes, customer replies, minutes, contracts. I find in a day what can be handed over and what it gives you back in hours, then put the first process under contract within a week: the written request, the sign-off, the register. Everything reads on one page, your desk: what is waiting for you, what went out without you, what nobody is allowed to guess. Law firms, agencies, hotels, support desks: the mechanism is the same.

Read the page: with code · without code · the desk, without code. Version française : avec du code · sans code · le bureau sans code.

Write: simon.pacifae.dupont@gmail.com · LinkedIn