Audit your workflow
Spend a day noting where your coding time actually goes, and flag the repo-wide grunt work autocomplete never touches: tracing a bug across ten files, rebuilding boilerplate, wiring the same API-plus-UI-plus-tests change by hand.
Learning Path
Engineering & Dev
Drive AI agents to do the multi-file work in your own repo, from scaffolding to debugging to shipping a full feature, then gate every AI-written change through code review and tests you actually read.
For working engineers13 modules~4h 54m total6 tools
Finish with: A CLAUDE.md in your repo plus a subagent-and-MCP setup, so every agent run starts from the same clean, repo-specific context
Who this path is for
Working software engineers and technical founders who ship code every day and want agents doing the repo-wide grunt work, not just autocompleting the next line.
You already have autocomplete, but the real time sinks are still manual: tracing a bug across ten files, building a feature that touches the API, the UI, and the tests, rewriting the same boilerplate again. Bolting a chat window onto that and copy-pasting snippets back and forth barely helps.
Build next: Ship one real feature end to end in your own repo, with an agent planning the change, editing across files, and running the tests before you review the diff.
Outcomes
What you'll be able to do
- A feature shipped end to end in your own repo by an agent, with the full diff reviewed and the test suite green before you committed
- A review-and-test gate on your pull requests: CodeRabbit or Greptile on the diff, plus Qodo behavioral tests on the functions you changed
- A CLAUDE.md and subagent setup that makes every agent run start from the same clean, repo-specific context
Curriculum
What you'll actually build
13 modules, ~4h 54m of hands-on work. Every lesson centers on one tool and ends with a deliverable you keep.
Draft functions from a comment with Copilot
A working function you wrote from a one-line comment, plus your own rule for when to accept or reject a suggestion
GitHub CopilotFree starter lesson14 min
Draft functions from a comment with Copilot
A working function you wrote from a one-line comment, plus your own rule for when to accept or reject a suggestion
GitHub CopilotFree starter lessonGitHub CopilotDev
Inline code suggestions in your editor. You work inside the real tool for all 14 minutes; the module ends when the deliverable above exists.
Refactor and explain code in Cursor's chat
A small multi-file refactor made through Cursor chat, with every diff read before you accepted it
Cursor18 min
Refactor and explain code in Cursor's chat
A small multi-file refactor made through Cursor chat, with every diff read before you accepted it
CursorCursorDev
AI-native code editor with multi-file edits. You work inside the real tool for all 18 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Scaffold an endpoint and its tests with Claude Code
A new API endpoint with passing tests, scaffolded by Claude Code from a tight prompt and the exact relevant files
Claude Code24 min
Scaffold an endpoint and its tests with Claude Code
A new API endpoint with passing tests, scaffolded by Claude Code from a tight prompt and the exact relevant files
Claude CodeClaude CodeDev
Agentic CLI for complex codebases. You work inside the real tool for all 24 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Debug from a failing test with Claude Code
A bug traced across files and fixed, with the previously red test now green and the fix diff read line by line
Claude Code26 min
Debug from a failing test with Claude Code
A bug traced across files and fixed, with the previously red test now green and the fix diff read line by line
Claude CodeClaude CodeDev
Agentic CLI for complex codebases. You work inside the real tool for all 26 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Ship a feature across API, UI, and tests with Cursor's agent
One feature shipped end to end in your own repo, the full diff reviewed and the suite passing before you committed
Cursor30 min
Ship a feature across API, UI, and tests with Cursor's agent
One feature shipped end to end in your own repo, the full diff reviewed and the suite passing before you committed
CursorCursorDev
AI-native code editor with multi-file edits. You work inside the real tool for all 30 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Generate a production React screen with v0
A React/Tailwind and shadcn screen generated in v0.app, dropped into a branch, and wired to your real backend
v020 min
Generate a production React screen with v0
A React/Tailwind and shadcn screen generated in v0.app, dropped into a branch, and wired to your real backend
v0v0Dev
Generate production UI from text prompts. You work inside the real tool for all 20 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Validate one flow fast in Bolt.new
A working browser prototype that proves one flow, plus a short list of the parts to rebuild and harden in your real repo
Bolt.new18 min
Validate one flow fast in Bolt.new
A working browser prototype that proves one flow, plus a short list of the parts to rebuild and harden in your real repo
Bolt.newBolt.newDev
Full-stack apps from a single prompt. You work inside the real tool for all 18 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Gate every pull request through CodeRabbit
CodeRabbit wired onto your pull requests, giving a structured first-pass review on every AI-written diff before a human looks
CodeRabbit18 min
Gate every pull request through CodeRabbit
CodeRabbit wired onto your pull requests, giving a structured first-pass review on every AI-written diff before a human looks
CodeRabbitCodeRabbitDev
AI reviews pull requests line by line and flags bugs in GitHub, your IDE, and the CLI. You work inside the real tool for all 18 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Catch cross-file bugs with Greptile review
A whole-codebase Greptile review on a real PR, flagging cross-file and convention breaks a diff-only tool would miss
Greptile20 min
Catch cross-file bugs with Greptile review
A whole-codebase Greptile review on a real PR, flagging cross-file and convention breaks a diff-only tool would miss
GreptileGreptileDev
AI code review agent that maps your whole codebase to catch bugs in pull requests. You work inside the real tool for all 20 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Close the test gap with Qodo
Behavioral unit tests generated by Qodo's /test for your changed functions, read to confirm they assert the intent you meant
Qodo22 min
Close the test gap with Qodo
Behavioral unit tests generated by Qodo's /test for your changed functions, read to confirm they assert the intent you meant
QodoQodoDev
AI agent that reviews code, generates tests, and enforces standards in the IDE and PRs. You work inside the real tool for all 22 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Delegate a long refactor to Codex
A multi-hour refactor run as a Codex cloud task, returned as a pull request with diff and logs you reviewed before merge
Codex26 min
Delegate a long refactor to Codex
A multi-hour refactor run as a Codex cloud task, returned as a pull request with diff and logs you reviewed before merge
CodexCodexDev
OpenAI cloud agent for parallel tasks. You work inside the real tool for all 26 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Orchestrate parallel agents in Google Antigravity
Several agents run at once from Antigravity's Manager view, checked against their Artifacts (plan, screenshots, logs) before you accept the work
Google Antigravity28 min
Orchestrate parallel agents in Google Antigravity
Several agents run at once from Antigravity's Manager view, checked against their Artifacts (plan, screenshots, logs) before you accept the work
Google AntigravityGoogle AntigravityDev
Agentic dev platform where AI agents plan, write, and verify code across editor and browser. You work inside the real tool for all 28 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Engineer your repo's context with CLAUDE.md and subagents
A CLAUDE.md in your repo plus a subagent-and-MCP setup, so every agent run starts from the same clean, repo-specific context
Claude CodeCapstone30 min
Engineer your repo's context with CLAUDE.md and subagents
A CLAUDE.md in your repo plus a subagent-and-MCP setup, so every agent run starts from the same clean, repo-specific context
Claude CodeCapstoneClaude CodeDev
Agentic CLI for complex codebases. You work inside the real tool for all 30 minutes; the module ends when the deliverable above exists.
This module unlocks with Pro. Checkout is not open yet; early access costs nothing today, no card.
Free gets you started today. Pro unlocks all 8 paths when checkout opens, no card and no charge until then. See what Pro unlocks
Start here
Your first three actions
- Turn on GitHub Copilot in your editor and let it write the next function from a comment describing what you need
- Point Claude Code at your repo and ask it to trace a bug across files, then approve the fix it proposes
- Describe one screen to v0 and drop the generated React component into a branch
Workflows you'll actually run
- Give Cursor a feature spec and use its agent to edit the API, the UI, and the tests in one pass, reviewing each change before you accept it
- Hand Claude Code a failing test and let it read the stack trace, edit across the codebase, and rerun the suite until it passes
- Kick off a multi-hour refactor as a background agent in Codex, keep working, then review the pull request it opens
Common traps
Where people go wrong
Accepting AI diffs you never actually read
Read every generated line before you commit and keep tests around the change; treat a GitHub Copilot or Cursor suggestion as a junior's draft to verify, not a merge-ready PR.
Collecting AI coding tools instead of building a workflow around one
Pick one tool like Claude Code or Cursor, learn to feed it tight context and tests, and judge it on your own repo instead of the next demo.
Your first week
Your 7-day starting plan
A practical first week to get one real workflow running. Not a full roadmap, a place to start.
Pick one task
From that list, pick the single most repetitive thing you do each week, like scaffolding a new endpoint with its tests, and make that the one task you hand to an agent first instead of trying to automate everything at once.
Build first workflow
Point Claude Code at your repo (or turn on GitHub Copilot and use Cursor's agent) and set it up to run that one task from a tight prompt plus the exact relevant files, not a snippet copy-pasted into a chat window.
Test on real work
Run the workflow on a real ticket you were going to do anyway, and read every generated line before you commit, treating each Cursor or Claude Code diff as a junior's draft to verify rather than a merge-ready PR.
Improve the quality
Where the output missed, fix the input: give it better context, more tests around the change, and a sharper spec, then rerun and compare the second diff against your first to see whether the setup or the model was the problem.
Write it down
Capture the version that worked as a short prompt-and-context checklist, or a CLAUDE.md in the repo, so the next run starts from the same setup instead of you reconstructing it from memory.
Decide what scales
Pick the next task worth handing off and match it to the tool: a long refactor to a background agent in Codex, a fresh UI screen to v0, then review the pull request or component before it lands.
Progress signals
How you know you're improving
- You read the agent's diff and it needs fewer corrections before you commit it.
- You reach for Claude Code or Cursor on a multi-file change without thinking, instead of doing it by hand first.
- You can hand off a whole feature across the API, UI, and tests and have the suite pass before you review, instead of babysitting each line.
Toolkit
Tools you'll master
Every tool here is a real product you can open today. Browse the full directory of 68 tools.
Explore other paths
Not sure this is your path?
The starter lesson is free, so open it and see. Or take the free assessment: it reads where your comeback stands and recommends the comeback course to start with.