Seven steps from curious to capable.
Not a syllabus — a path. Each step is something to do, not just read. The 'proof' for each is what tells you you're ready for the next.
Have a real conversation with a model
Open Claude or ChatGPT and use it for a real task today — a draft, a tricky email, an unfamiliar code error. Notice where it shines and where it stumbles. This shapes your intuition more than any tutorial.
Learn prompt engineering by feedback
Write a prompt. Read the output. Find one thing that's wrong. Edit the prompt to fix it. Repeat. After ten rounds you'll have internalized more than any course will teach.
Use AI in your daily workflow
Install Claude Code or Cursor. Use it for a week of real work. The skill of collaborating with an agent — what to delegate, what to keep — only develops by doing.
Build something small with the API
Pick a tiny project — summarize your inbox, tag your photos, draft replies. Call the API directly. You'll learn more from 50 lines of glue code than from any framework tutorial.
Add retrieval (RAG)
Give your model access to your own documents. Chunk them, embed them, search them, paste the top results into the prompt. This unlocks 80% of useful business apps.
Add tools and a loop
Let the model call functions — read a file, hit an API, run a query — and let it loop until done. Congratulations: you've built an agent.
Add evals before you scale
Write 20 test cases with expected behaviors. Run them every time you change a prompt. Without this, every 'improvement' is a guess.
Pick a tool and start.
The shortest path through this guide is to pick one chat assistant and one coding tool, then use them every day this week.