New: Roadmaps ordered paths through our cheat sheets and flashcards, so you always know what to study next.
Explore themSee what's new on GitHubFrom your first AI-generated suggestion to a reviewed, git-backed workflow you'd trust in production.
A 12-step learning path. Follow it in order, or jump to what you need.
This path is for developers who already write code and want to get fast, reliable output from AI coding assistants and autonomous agents without losing control of their codebase. Expect about 4 to 5 weeks at a few hours a week, moving from turning on an assistant and vibe-coding a real feature, through handing work to autonomous agents, into the review and guardrail habits that keep AI-generated code honest. It does not cover building or fine-tuning the language models themselves, that's what Generative AI & LLM Engineering already owns; here the model is a given tool, and the skill is directing, testing, and reviewing what it produces. By the end you can prompt and steer an AI coding agent through a real feature, catch and fix what it gets wrong before it ships, and run a workflow disciplined enough to trust with production code.
Expected: comfort writing and running code in at least one programming language. Helpful but not required: prior use of Git or an AI coding assistant like GitHub Copilot.
Open your editor and switch on an AI coding assistant today: this path starts wherever a working suggestion first appears on your screen, no other setup required.
Commit before every AI-assisted change so a bad suggestion from step 1 is one revert away, not a rewrite of an afternoon's work.
Combine the prompting from step 1 with the commit habit from step 2 to carry a real feature from idea to working code in one sitting, rules files and all.
You can turn on an AI assistant, keep a git safety net under it, and carry a real feature from idea to working code in one sitting. Next up: keeping that fast output honest with tests.
Finish this section to unlock.
+100 XP
Write the failing test before you ask the AI to write the code, so step 3's fast output has a spec keeping it honest instead of quietly drifting.
Hand a whole task, not a single suggestion, to an autonomous agent: this is the shift from typing alongside AI to supervising it, and it rattles most developers the first week.
Read what the agent from step 5 handed back the way you'd read a stranger's pull request; expect to catch yourself rubber-stamping confident-looking code before real verification becomes the habit.
You can hand a real task to an autonomous coding agent and review what it hands back like a professional, and a few minutes on due flashcards from chapter 1 keeps your git habits fresh while you take this on. Next up: guardrails that work even when nobody's watching.
Finish this section to unlock.
+100 XP
Wire static analysis and coverage checks into your pipeline so the review habit from step 6 keeps working even on the commits nobody reads line by line.
Readers curious why an assistant sometimes invents a library that doesn't exist can trace it here, straight back to the sampling choices behind step 5's agent.
Developers standardized on GitHub Copilot should go deep here for its slash commands, chat participants, and agent mode beyond what step 1 introduced.
Developers who picked Cursor as their daily driver get its agentic workflows and composer features here, the Cursor-specific half of what step 3 practiced.
Developers running Claude Code from the terminal get its hooks, skills, and subagent patterns here, the most autonomous version of the agent workflow from step 5.
Pull prompting, git discipline, tests, agents, review, and guardrails into one repeatable workflow you can run on your next real project, from a rough idea to something a team can actually ship.
You can prompt and steer an AI coding agent through real work, catch what it gets wrong before it ships, and run a workflow disciplined enough for production, earning your AI-Assisted Coding Practitioner badge.
Finish this section to unlock.
+100 XP
Finish every required step, at least 70% of them genuinely done (not skipped), to earn this badge and 500 XP.