← sources

When coding agents work autonomously, what do you actually trust them to do?

source: reddit (reddit:AI_Agents)published 2026-10-20 09:57ingested 2026-10-21 02:35

retained text

I've been thinking about what happens when we move from one coding agent to several working across the same codebase.

Writing code is becoming less of a bottleneck. Coordinating agents, understanding what changed, and deciding when a human should step in seem like much harder problems.

Some people use GitHub issues and PRs, others build custom orchestration scripts, retrospectives, or permission wrappers.

I'm curious about people actually using these workflows:

* How do you decide what an agent can do without asking you?
* Can you reliably reconstruct what happened when a run fails?
* What still requires your attention, even with a good setup?

I'm especially interested in what has **failed in practice**, not what a perfect agent workflow should look like.

document fetched 2026-10-21 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xanzxx/when_coding_agents_work_autonomously_what_do_you