lemon Lesson 1 of 7 · 3 minutes

From answers to actions

See what a coding agent can do inside a project, and what still belongs to you.

You already know the basic interaction: describe what you want, read the AI’s answer, and decide what to do with it. A coding agent adds a workspace and tools to that conversation.

It can inspect the files in a project, make edits, run commands, and show you the result. That saves you from carrying every fragment between the chat and your editor. It also makes a vague request more consequential.

In a chatWith a coding agent
You paste in the contextThe agent inspects the project
The AI suggests an answerThe agent can edit the real files
You run the commandsThe agent can run checks for you
You carry the answer outThe result already exists in the workspace

Your job does not disappear

The agent can do the mechanical work. You still decide:

  1. What outcome matters.
  2. What it may change.
  3. Which decisions require your judgment.
  4. What evidence would make the result trustworthy.

That division is the foundation of this course. You are not learning a magic prompt. You are learning a small working loop: scope, plan, act, verify, review.

How the agent acts

Most coding agents work through the same interfaces programmers and scripts already use:

  • A CLI is a program the agent runs as a terminal command. It is often the clearest route for files, Git, builds, tests, and local tools.
  • An API lets code send structured requests to a service. The agent can use it directly or through a small command that handles details such as authentication and pagination.
  • MCP gives an agent host a standard way to discover tools and live context. It is useful for shared service connections, but it is not a replacement for every CLI or API.

You do not need to operate all of these yourself. The important habit is to prefer a clear, inspectable tool surface over repeated clicking or improvised steps. The field manual has a fuller decision guide in Choose the right tool surface.

The safest first move is often read-only: ask the agent to explain the project before asking it to change the project.

Try it

Make the call

Pick an option to see how the rest of the session plays out.

agent

Connected to your project. I can read the files, edit them, and run commands like the build. What would you like to do?

Which request makes good use of an agent without giving away the important decisions?

agent

Done. I scaffolded a new app with a landing page, a pricing page, and a blog, picked a color scheme, and rewrote your homepage copy to match.

you

Wait. This doesn’t look like my site anymore. Which of these decisions were mine to make?

not yet

This delegates the product decisions as well as the typing. There is no shared picture of done, so a confident result can still be the wrong product.

agent

It’s a template site with three pages. The homepage text lives in one file, the styles in another, and `npm run build` checks that everything still compiles. Two things stood out: one image is oversized, and a link on the About page is broken. I haven’t changed anything.

you

Now I can see two small, safe tasks worth trying, and I know how to check them.

good call

This uses the agent’s access to the project while keeping the first move read-only. You get orientation before authorizing change.

agent

In general: HTML holds the content, CSS controls the appearance, and JavaScript adds behavior. Want me to look at how your project does these things specifically?

you

Good explanation, and I could have gotten it from any chat. I still haven’t learned anything about my own site.

close, but

That is a useful chat question, but it does not use the agent’s ability to inspect your actual project.

Take it to your tool

Try this in your agent

With the project open in your coding agent, paste:

Before doing anything, inspect this project and tell me: what it appears to be, how it is organized, which project instructions you read, and what CLI commands seem to check that it works. Do not edit files yet.
Got the idea?

Mark it complete on this device. You can return or reset the course at any time.