A software engineering approach where, instead of directly prompting an AI coding agent turn by turn, a developer designs an automated system (the loop) that discovers work, delegates tasks to agents, checks results, persists state, and decides the next action on a schedule or until a goal condition is met.
So what? The leverage point shifts from writing good prompts to designing systems that do the prompting, which changes the nature of the engineer's role from operator to architect.
Example: A morning automation that reads yesterday's CI failures and open issues, spawns sub-agents to draft fixes, runs a verifier against the results, opens a PR, and updates the issue tracker, all without the engineer typing a single prompt.
Related term: Pilot Agent.
So what? The leverage point shifts from writing good prompts to designing systems that do the prompting, which changes the nature of the engineer's role from operator to architect.
Example: A morning automation that reads yesterday's CI failures and open issues, spawns sub-agents to draft fixes, runs a verifier against the results, opens a PR, and updates the issue tracker, all without the engineer typing a single prompt.
Related term: Pilot Agent.
