The Landscape & Setup
Coding agents come in a few basic shapes. You do not need all of them — you need to pick one and set your project up so it can actually help. This lesson does both.
The Landscape
Most coding agents fall into one of three groups. They overlap, and many people use more than one.
- Terminal / CLI agents — run in your shell, work across the whole project (e.g. Claude Code).
- IDE-integrated assistants — live inside your editor, close to the file you are on (e.g. Cursor and editor assistants).
- Web / hosted agents — run in the browser or the cloud, handy when you are away from your machine.
Setup That Helps Any Agent
The setup is the same whichever tool you pick. Three things make an agent far more useful before you type a single request.
A Good Instructions File
Writing this one file is the single highest-leverage setup step. Keep it short and concrete.
- Conventions — how the code is organised and named.
- How to run and test — the exact commands.
- What NOT to touch — files, secrets, and areas to leave alone.
Fit It to Your Workflow
Terminal versus editor is a matter of preference — pick what feels natural. The concepts carry across every tool: giving good context, reviewing what the agent produces, and iterating. Those matter far more than the brand.
Build It
How to implement: create a one-page instructions file for your project — how to run it, its conventions, and its don'ts.
- Weekly AI Tasks tracker — a rules file stating the stack, how to run the server, and “never commit secrets” keeps the agent on track.
- Personal brand site — note the tone, the sections, and “every claim must trace to real material” so generated copy stays honest.
What you learned
Coding agents come in three shapes — terminal, IDE-integrated, and web / hosted. Whichever you choose, a clean repo, a README, and a short instructions file are what make it genuinely helpful.