What a Coding Agent Actually Does
A coding agent is not a search box and it is not autocomplete. It is a program that wraps a language model and gives it hands: the ability to read your files, run commands in your terminal, edit code, and check whether the result worked. You describe a goal in plain language, and the agent works through it step by step until the job is done or it gets stuck.
The thing that makes it feel like magic is the loop. The agent reads context, takes one action, looks at what happened, then decides the next action. Tools like Claude Code, Cursor, and Windsurf all run some version of this loop. Once you see it, the rest of the course is mostly learning how to feed it well.
The four steps of the loop
- Read: the agent opens the files and runs searches to understand your project.
- Act: it edits a file or runs a command.
- Check: it runs the test or reads the output to see if it worked.
- Repeat: it uses what it learned to pick the next move, or stops when finished.
Chat vs agent, side by side
| Chatbot | Coding agent | |
|---|---|---|
| Scope | Answers one message | Works a whole task |
| Tools | Usually none | Files, shell, search |
| Who runs the code | You copy-paste it | The agent runs it |
| Best for | Questions | Shipping changes |
A first session, plain and simple
Notice there was no clever wording. A clear goal plus a way to verify the result is the whole foundation. Models in 2026 such as Claude Opus 4.8 and Claude Sonnet 4.6 are strong enough that your job shifts from typing code to steering and checking it.