Loops in the wild

The same shape shows up everywhere.

A coding agent edits, runs tests, reads failures, edits again, and stops when the checks pass or the budget runs out. Claude Code, Codex, and similar tools feel different at the surface, but the engine is the same repeated decide, act, check cycle.

A research agent searches, reads, writes notes, decides what is still unknown, searches again, and stops when the question is answered well enough. The check might be a source requirement, a contradiction scan, or a human review of the final synthesis.

A monitor wakes on a schedule, checks sources, compares with last time, reports only what changed, and stops. The schedule is the trigger. The loop is what the agent does once it wakes up: read, compare, decide, report, and stop.

What varies is the trigger, the tools, and the stop condition. What stays the same is the cycle and the need for checks.

Sources: Simon Willison, How coding agents work; Addy Osmani, Loop Engineering.