
How Ashlr runs on AI agents, part 2: how agents ship code
Agents can write code all night. The hard part is making sure bad code never reaches your users. These are the rules our agents follow to ship.
The field note
Letting an agent write code is easy. Letting it ship code is a trust problem. Our engineering team is mostly agents, and they work overnight without anyone watching. So the rules matter more than the model.
Rule one: nothing merges without green checks. Every change goes through a pull request. The automated tests and checks have to pass. If they fail, the change waits.
Rule two: one pull request at a time. Agents do not pile changes on top of each other. Each one lands, gets verified, and only then does the next start. It is slower on paper and faster in practice, because problems stay small and easy to trace.




