Lesson 1 of 6

How a coding agent works

7 min read

You type 'fix the failing test' and walk away. A minute later it's green. Between those two moments, the agent did five things over and over. What were they?

One prompt, many turns

A coding agent isn't a single call to a model. It's a loop. You give it a goal — not a script of steps — and it works toward that goal one turn at a time: read some code, decide what to change, make the change, run a tool to check itself, look at the result, and go again. It keeps looping until the goal is met or it runs out of road.

A coding agent is a model wrapped in a loop with tools. Give it a goal and it plans, acts, and observes — over and over. That loop is what makes it an agent, not just a chatbot.

The loop reacts

The power isn't any single step — it's that the agent reacts to what it observes. The first edit didn't fully work: a test still failed on a blank input. So the agent didn't stop — it read the failure, re-planned, and tried again. That feedback loop, running against real tools, is why an agent can finish a job a one-shot answer can't.

Because it observes the result of each action, the agent can course-correct. A failing test isn't the end — it's the next turn's input.

This is why coding agents feel slow but thorough: each loop is a real round-trip — read a file, run the tests, look at the output. You're watching it work, not just type.