Track

AI for Developers

Using AI on real code without shipping things you do not understand. Every item enforces reading and verifying before accepting.

Who it's for: Developers already writing code who want AI to help rather than to guess.

By the end you'll have

A debugging loop that diagnoses before it fixes
The habit of reviewing generated code as if it came from a stranger
Refactoring backed by characterisation tests

Why this order

Prompting fundamentals before anything agentic, because chaining unreliable steps multiplies the unreliability rather than averaging it. A developer who skips straight to agents ends up debugging a system where every step is slightly wrong and no individual step is obviously broken.

Retrieval comes before agents for the same reason: an agent that cannot reliably ground itself in your data is an agent that confidently makes things up at every step.

How long it takes

Ten to fifteen hours if you build everything rather than reading it. The projects assume you write code and will take longer than the reading suggests, which is the point.

How you'll know you're done

You have finished when you have something running that uses more than one model call, handles a failure in the middle without falling over, and that you can evaluate — meaning you could tell whether a change made it better. Without that last part you have a demo rather than a system.

Not sure this is the right track?

There are 4 tracks and 24 standalone projects and challenges.

Browse everything