Loop Engineering
Loop engineering is the practice of designing the repeatable control system around an AI coding agent — goal, context, tools, observation, verification, retry policy, and stop rules — instead of optimizing a single prompt.
Detailed definition
For AI coding agents, loop engineering replaces task-by-task babysitting with a durable Plan → Act → Observe → Verify → Stop cycle. The agent is given a bounded goal, reads repository evidence (test output, compiler errors, diffs, logs) before its next step, retries only with a changed strategy, and stops on token or cost caps, repeated failure, wider-permission requests, or a required human checkpoint. This makes agent work auditable: every retry has a reason, every escalation has an owner, and every completed run leaves a concise artifact. Loop engineering is distinct from prompt engineering, which improves one instruction, and from a cron job, which runs a fixed command on a schedule without observing state.