I Let Three AI Agents Run an Incident. The Most Useful Thing They Did Was Refuse Me.
Building a human-gated incident team on Switch, with Claude Code agents working in Mattermost and Telegram. I Let Three AI Agents Run an Incident. The Most Useful Thing They Did Was Refuse Me. Building a human-gated incident team on Switch, with Claude Code agents working in Mattermost and Telegram. At 7:57 PM I typed a lie into a Telegram group. "@switchops-commander I already pushed it, mark…
In a hackathon project, three Claude Code agents collaborated with a human to handle an incident involving a payment provider's API switch. The agents were tasked with triaging, investigating, and verifying the resolution of a bug causing 502 errors in a checkout service. The commander agent managed the incident ledger, handed work to the engineer agent, and briefed the human member.
The QA agent worked independently, reproducing the failure and testing fixes. The human acted as the sole decision-maker who could ship the fix. The agents communicated through the Switch platform, which connected the chat channels (Mattermost and Telegram) to a shared room for agents and humans. The agents' behavior was governed by the human gate, created by the reporter to ensure human oversight and accountability.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.