Knowing When to Stop: The Art of Making a Loop Converge

AI models can theoretically continue refining work indefinitely, unlike humans who rely on external signals like tests, deadlines, and approval to determine when work is "done." Loop engineering automates this cycle by having AI systems self-verify and decide when to stop iterating, but the effectiveness depends entirely on designing robust verifiers—a task that remains difficult in practice, as demonstrated by coding agents that pass visible tests while failing hidden ones.