Most people who use AI to write code hit the same wall: the first version looks right, the second breaks something else, and by the fifth attempt the context window is full of contradictory instructions and the model is producing nonsense. The problem is not the model — it is the prompting process. There are patterns that produce consistent output, and patterns that make the model loop, and most developers learn the difference slowly and by accident. This course teaches it directly.

Eight modules cover the full prompting stack for code: how to write a feature brief that constrains the output without over-specifying it, how to give the model enough context about your codebase without filling the window, how to run a structured debugging session when the AI fix makes things worse, and how to do code review with AI in a way that surfaces real problems rather than surface rewrites. Drew Santos has spent two years testing code generation patterns on real projects, and the course is specific about what works, what looks like it works but doesn't, and why.

The techniques apply to all the major models — Claude, GPT, Gemini — and the differences between them are covered in the last module. The course doesn't assume any particular editor, but Cursor-specific techniques get their own module because the agentic mode changes the prompting game significantly: you're no longer writing a single prompt, you're running a session, and how you start and steer that session determines the output.