monitor-app
built · first live drill pending
When production breaks, an AI investigates in the repo and drafts the fix PR. It can't merge and it can't send — by construction.
Production errors become messages on a durable queue; a dispatcher hands each to an AI session running inside the affected project's own repository, which investigates and drafts a pull request. The human gates are structural: no merge rights, and no email-send capability exists at all.
The interesting decision in this ops pipeline isn't the AI — it's the gates. "The agent sends after approval" was rejected deliberately: a gate you must remember to keep closed will eventually be left open; a gate that does not exist as a capability cannot be. So the triage agent drafts PRs it cannot merge and replies it has no tool to send. This is the pattern I bring to client AI work: give the machine the legwork, make the dangerous actions structurally impossible, and let humans spend their attention on judgment instead of investigation.
Want AI in your ops without AI in your blast radius?
Let’s talk