Review Glossary FR

Size

Eyes

Finish a part in any dojo to unlock the next eyes.

Work well with AI models, ten minutes a day.

One chapter a day, checked against the docs and real code. Questions you get right come back 1, 3, 7, 21 and 45 days later.

Today · day 1 · 10 min

Tokens
Start reading →

0 of 40 chapters done

How models work: 0 of 8Prompting: 0 of 8Tools and agents: 0 of 8Claude Code: 0 of 8Building with the API: 0 of 8
One pillar per part. It fills as you finish its chapters.

The path

How models work: 0 of 8How models work

  1. Tokens 1
  2. Predicting the next token 2
  3. The context window 3
  4. Embeddings 4
  5. Confident and wrong 5
  6. How a model is trained 6
  7. Choosing a model 7
  8. Cost and latency 8

Prompting: 0 of 8Prompting

  1. Clear instructions 9
  2. Showing examples 10
  3. System prompts and roles 11
  4. Structuring a prompt with XML tags 12
  5. Letting the model think 13
  6. Getting JSON back 14
  7. Working with long documents 15
  8. Prompt injection 16

Tools and agents: 0 of 8Tools and agents

  1. Tool use 17
  2. The agent loop 18
  3. Model Context Protocol 19
  4. Retrieval-augmented generation 20
  5. Memory across sessions 21
  6. Splitting work across agents 22
  7. Testing a prompt 23
  8. Keeping an agent safe 24

Claude Code: 0 of 8Claude Code

  1. Claude Code in one session 25
  2. CLAUDE.md 26
  3. Permission modes 27
  4. Hooks 28
  5. Skills 29
  6. Slash commands and plugins 30
  7. Subagents in Claude Code 31
  8. Claude Code in scripts and CI 32

Building with the API: 0 of 8Building with the API

  1. The Messages API 33
  2. Prompt caching 34
  3. Streaming 35
  4. Batch processing 36
  5. Grading with a model 37
  6. Reviewing code an AI wrote 38
  7. When not to use a model 39
  8. Capstone: a small coding agent 40