Agent Loop
The repeated cycle where an agent reasons, calls a tool, observes the result, and decides the next step.
Plain English
It is the heartbeat of an agent: decide, act, look at what happened, repeat.
Example
Search the docs, read a result, update the plan, run a command, inspect the error, and try a fix.
Why it matters
Loops make AI useful for open-ended work, but they also create compounding cost and error risk.