← All questions
Should I interrupt an AI agent while it is working?
Usually not. Mid-stream prompting tends to leave half-applied patterns, contradictory constraints and broken invariants. Let the run finish, then review a clean diff.
Chat makes everything look interruptible, but software work depends on continuity. An engineer halfway through a refactor holds a temporary model of old and new structure, and so does the agent.
Constant interruption also keeps you in perpetual supervision and means you never get a clean diff to review.