Learn the harness, not just the model.
Tutorials, recipes and reference for the tools you point at your codebases. Take the course if you want to level up, or just raid the reference if you're already an expert ;)
- 62 LESSONS
- 6 HARNESSES
Pick your go-to harness and lesson one opens. The course itself is universal, your pick just sets the flavour; you can still change it mid-read.
- Claude CodeAnthropicSkills, subagents, hooks, sandboxing, plugins, the Agent SDKUSE THIS →
- Codex CLIOpenAISkills, approvals, sandbox modes, exec, four surfacesUSE THIS →
- GitHub Copilot CLIGitHubSkills, the async coding agent, code review, org distributionUSE THIS →
- CursorAnysphereRule scoping, Cloud Agents, hooks, cursor-agentUSE THIS →
- Antigravity CLIGoogleExtensions, long-context work, and the Gemini CLI legacyUSE THIS →
- Kimi Code CLIMoonshotModel routing, the rebuild, and what it has no sandbox forUSE THIS →
- Not sure yetSee the same job in all five, side by side, and pick from that.HARNESS COMPARISON →
Start from the beginning or skip to what sounds interesting. After all, what's the point in trying to learn what you already know, right?
Driving the harnessThe transferable half. Everything here is true of every harness — the switcher swaps the part that isn't.- 01FoundationsWhere the model ends and the harness begins. Every other chapter assumes this one.
- 02Prompting an agent, not a chatbotThe habits that transfer from chat are the ones that hurt most here.
- 03Commands and skillsOne mechanism, two names — and an open standard several vendors now share.
- 04Tools and MCPTools, not prompts. Wiring a model into the systems it has to touch.
- 05Subagents and fan-outParallelism you can still review.
- 06Hooks and guardrailsDeterminism around a non-deterministic core. At this speed, the gate is the design.
- 07Cost and controlThe loudest pain in the field, and the one the two chapters above create.
- 08Headless and CIThe same harness with nobody at the keyboard.
- 09Doing this with other peopleEverything above, times a team.
We're rolling out supplements for the harness-specific parts too. Stay tuned for Claude Code.
Three subjects so big they deserve their own modules. Past the basics and into power-user territory.
- Context engineeringThe window is a budget. This is how you spend it — and why everything gets worse the longer you go.
- Reviewing agent outputThe skill nobody had to have before, and the one that decides whether any of this is safe to ship.
- Security for AI-assisted teamsThe threat model nobody wrote down, and the answers your security team will ask for.
The advanced features your harness has and the others don't. Stay tuned.
- Claude Code, specificallyThe surfaces no other harness has an equivalent for.
- Codex CLI, specificallyWhere it diverges from everything else — which is mostly in how many places it runs.
- GitHub Copilot CLI, specificallyThe one your company already pays for.
- Cursor, specificallyRule scoping, cloud agents, and the CLI most Cursor users never open.
- Antigravity CLI, specificallyGoogle's current agentic CLI: extensions, a very long window, and what changed from Gemini CLI.
- Kimi Code CLI, specificallyThe rebuild, the routing, and the one thing it deliberately doesn't have.
If you're no story-listener and would rather skim an example and get on with it, start here and forget we ever mentioned lessons. Pick your problem: “I need to…
- …stop re-explaining the same architecture every morning”AGENTS.MD, CLAUDE.MD AND THE REST: WHAT BELONGS IN A MEMORY FILE →
- …stop it burning through my budget overnight”BUDGETS, CAPS AND A KILL-SWITCH FOR UNATTENDED RUNS →
- …keep it away from my migrations”BLOCKING THE COMMIT THE AGENT SHOULDN'T MAKE →
- …stop it asking permission for every single thing”PERMISSION FATIGUE AND THE MIDDLE PATH →
- …undo what it just did to thirty files”GIT AS THE REAL UNDO →
- …turn a habit I keep typing into one command”SETTING /GOALS →
- …get a plan out of it before a single file changes”ASKING FOR A PLAN YOU CAN ACTUALLY REJECT →
- …hand a forty-file refactor to five agents and still review it”WRITING A SUBAGENT BRIEF THAT RETURNS DATA, NOT PROSE →
- …give it database access without giving it my database”SCOPING AND AUTH: WHAT A SERVER SHOULD REFUSE TO DO →
- …work out why it gets dumber the longer I go”CONTEXT ROT: WHY IT GETS WORSE THE LONGER YOU GO →
- …review a pull request I didn't write and can't skim”READING A LARGE DIFF YOU DIDN'T WRITE →
- …run the whole thing in CI with nobody at the keyboard”HEADLESS MODE: WHAT CHANGES WHEN THERE'S NO ONE TO ASK →
- …find out why my config is being silently ignored”VERIFY IT ACTUALLY LOADED: CONFIG THAT FAILS SILENTLY →
Some useful bits, for when the reading gets heavy. The glossary covers any term you don't know.
AI-native engineering keeps growing, and so does Academy. If you're missing something you'd like to read – say hi.