← sources

What broke first when your AI agent started running unattended?

source: reddit (reddit:AI_Agents)published 2026-10-12 11:01ingested 2026-10-13 02:36

retained text

I keep seeing the same pattern with AI agents. They work fine in a demo, someone leaves them running on a schedule or in production, and a week later someone finds them stuck in a loop or silently doing the wrong thing.

In my experience the first thing to break is almost never the model. It is the assumptions nobody wrote down. The prompt quietly assumes the input file will always be under a few hundred lines, or that the API returns the same fields it did last month. The agent handles the case it was described and skips the one nobody mentioned, without saying anything.

The second thing is state. An agent that runs for ten minutes can hold its whole plan in context. An agent that runs for hours, or wakes up on a schedule, starts making decisions from half-remembered state. It retries a step it already finished, or skips a step because it believes it did it in a previous run. That is how you get duplicate emails sent or duplicate orders placed, the whole embarrassing category.

What I am curious about: if you have run agents unattended for real, what broke first for you? A tool timeout that reported success? State drifting between runs? Some third thing I have not hit yet? And what was the fix that actually stuck?

document fetched 2026-10-13 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x3yw38/what_broke_first_when_your_ai_agent_started