The agentic loop (plan, act, observe)

What it is

An agentic loop is the repeated cycle an agent runs through: gather context, decide on an action, execute it (often via a tool), observe the result, and decide whether to continue or stop. The loop repeats until the model judges the task complete or a limit (turns, cost, time) is hit.

Why it matters for you

The loop is where cost and risk accumulate in production: each iteration burns tokens and time, and a poorly bounded loop can spin far longer than expected. Understanding this helps you ask the right questions when scoping an agent project — how many loop iterations is normal for this task, and what stops it?

Common failure mode

Without a hard stop condition (max iterations, budget cap, or explicit success check), an agent can loop indefinitely on a task it can't complete, silently burning API spend — the fix is always to bound the loop explicitly rather than trust the model to know when to give up.

This is exactly what we work through in a session.