7
Module 7
AI tooling
Local AI in VSCode for everyday work. Claude Code and the harness for spec-shaped work.
Module roadmap
2 chapters · 34 lessons
Local AI in VSCode
Ch 2217 lessons
Self-contained AI in your editor. Ollama, Continue.dev, the model landscape, quantisation. When local is enough, when you reach for a frontier model.
- 22.0Orientation: a coding model on your own laptop
- 22.1Why local matters
- 22.2Ollama: the local model runtime
- 22.3Installing Ollama on macOS, Windows, Linux
- 22.4Pulling your first model
- 22.5Running a model from the terminal
- 22.6Quantisation explained for users
- 22.7Hardware reality check
- 22.8The current coding-model families
- 22.9Continue.dev: the local-first VSCode extension
- 22.10Continue.dev configuration
- 22.11Inline edits with a local model
- 22.12Codebase indexing and retrieval
- 22.13Cline as the local agent-style alternative
- 22.14When local fails — the honest threshold
- 22.15The hybrid workflow
- 22.assignAssignment: Refactor a module with a local model only
Claude Code and the harness
Ch 2317 lessons
The CLI loop, /commands, skills, hooks, memory, subagents, plan mode, loop mode. Writing specs, not snippets.
- 23.0Orientation: what this chapter changes
- 23.1What 'the harness' actually is
- 23.2Installing Claude Code, first run
- 23.3The CLI loop: input → tools → output → next turn
- 23.4Permission modes and why they matter
- 23.5Slash commands and skills
- 23.6Memory: persistent notes the harness keeps for you
- 23.7Hooks: hard to grasp on day one, then indispensable
- 23.8Subagents and parallel work
- 23.9Plan mode: think before doing
- 23.10Loop mode: long-running self-paced work
- 23.11MCP servers in one sentence each
- 23.12When Claude Code beats your IDE and when it doesn't
- 23.13How specs become code: writing prompts not snippets
- 23.14Cost honesty: tokens, model tiers, when to drop down
- 23.15Versus Cursor / Windsurf / Aider — quick comparison
- 23.16Assignment: Ship a full feature end to end with Claude Code