# How do you decide when a coding agent has spent enough time on a task? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/d2944138449dde43

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-14 17:24 ingested 2026-10-15 02:37

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1x5xbqu/how_do_you_decide_when_a_coding_agent_has_spent/)

## retained text



```
I've had coding agents run for hours and finish with a confident account of work that didn't match what they'd actually done. Later edits were based on an increasingly stale picture of earlier ones, and leaving the agent running longer didn't fix that.



On OmniNode, which I work on, we've been breaking work into smaller tasks and starting each worker with the files and instructions for that task. That makes a failed attempt easier to inspect and rerun. It doesn't answer how long to let that attempt keep trying before giving it back to a person.



A time limit is straightforward, but it can cut off useful work. Tool activity is a poor substitute: an agent can read files and rerun commands for a long time without getting closer to the requested result. I'm more interested in whether it has produced something new that can be checked.



What do you use to make that call? If you run agents unattended, what specific signal makes you stop a task rather than give it another attempt?
```



document fetched 2026-10-15 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x5xbqu/how_do_you_decide_when_a_coding_agent_has_spent
