a practical guide

Getting the Most
out of Claude Code

(Claude Code)

Not a feature tour — the habits that make an AI coding agent actually productive, in about half an hour.

A free, practical guide to working well with Claude Code: why an agent is nothing like a chatbot, how to feed it the context it needs, planning before it builds, insisting on proof that the code works, and the setup — CLAUDE.md, custom commands, subagents, hooks, MCP — that pays you back every session.

in the real world

Agentic coding tools are becoming standard equipment for developers. The gap between people who love them and people who fight them isn't the tool — it's a handful of habits, and they're all learnable in an afternoon.

Start with the mental model →

Part 1 — The right mental model

1

It's an agent, not a chatbot

The mental shift that changes everything: you're delegating a task to something with hands, not asking a question.

2

Context beats cleverness

It starts every session knowing nothing about your repo. Feeding it the right context is the whole game.

Part 2 — The working loop

3

Plan before it builds

The cheapest place to catch a wrong approach is before a single file changes — so make it plan first.

4

Make it prove it works

"Done" is a claim, not evidence. Insist on tests, output and diffs — and use git as your undo button.

Part 3 — Make it yours

5

Teach it your project once

CLAUDE.md, custom commands, subagents, hooks and MCP — setup that pays you back every single session.

6

The habits that make it click

Steer early, keep tasks small, pick the right model — and know the moments to close the laptop and write it yourself.