12 min · up to 60 XP

Your AI toolkit

Chat assistants, in-editor copilots, and coding agents: what each is for and which to start with.

Sam made the classic mistake: she googled "best AI coding tool 2026". Forty tabs later she's drowning in names (ChatGPT, Claude, Copilot, Cursor, Claude Code) plus a comment thread of developers arguing with the confidence of people who have never been wrong in their lives.

In recruitment she'd know what to do with a shortlist this messy: stop reading opinions, start sorting by role. Underneath the forty tabs there are only three species of tool. Before any definitions, try sorting one job yourself:

Tap the diagram

Sam wants to hand over one whole task ("rename this project's colour scheme everywhere and check nothing broke") and have the tool find the files, make the edits, and run the checks itself. Tap the species built for that.

Meet the three properly. Chat assistants (ChatGPT, Claude, in your browser): you type, it answers, you copy-paste what's useful. That friction is a feature: every line passes through your eyes, so the read step is built in. It's where Sam has worked since lesson one.

Copilots (GitHub Copilot, Cursor) live inside your editor, suggesting the next lines as you type: speed, at the cost of making it easy to accept code you haven't read. Agents (Claude Code) take a described task and do the work across your project. Sam, sensibly, asks her chat assistant about the competition:

Predict the response

Sam does the most sensible thing imaginable: she asks her chat assistant about species two. Predict the reply.

What does an editor copilot actually do that you don't? Should I be using one instead of you?

What do you think the AI will do?

So which should you use? The trajectory this course follows: start in chat: zero setup, maximum learning per minute, home base for the next few modules. Add a copilot when you're building real projects, once reading is a habit strong enough to survive one-keypress suggestions. Meet the agents when you're comfortable describing tasks precisely and reviewing what comes back.

And the quiet news that makes the forty-tab argument irrelevant: it's the same LLM idea underneath all three. The skills you're building (describing clearly, reading carefully) transfer completely. You can start today, free, in a browser tab.

Fill the gaps

Fix the trajectory in your mind. Complete Sam's plan:

Start with a : zero setup, maximum learning. Add an when building real projects, and meet the agents when comfortable. Underneath, all three are , so your skills transfer between them.

Key takeaways

Every AI coding tool is one of three species: you sorted a real task before we'd defined a single one.
Chat assistants (browser, copy-paste) are the best place to learn: every line passes through your eyes.
Copilots suggest code inside your editor as you type; agents take a described task and navigate, edit, and test across your project.
The trajectory: start in chat, add a copilot for real projects, meet agents when comfortable, and you can start today, free, in a browser tab.
It's the same LLM idea underneath all three: describing well and reading carefully transfer everywhere.