When do you stop correcting an agent and start the task over?
retained text
Suppose you ask an agent to plan a weekend trip with a fixed budget, no rental car and some downtime each day. The first plan is too packed. You ask it to slow things down, and the new version looks better, but now one of the stops needs a car. You point that out, and the next version drops the budget breakdown. Each correction fixes something, but you still have to check whether the earlier requirements survived. At that point, another correction might be quicker than starting over. A fresh task might help, but it also means explaining everything again. What tells you it is time to restart rather than keep correcting? A concrete example of that turning point would be useful.
document fetched 2026-10-28 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xgz8vp/when_do_you_stop_correcting_an_agent_and_start