“I don’t really do prompt engineering anymore. There’s a loop, and the loop is prompting Claude and deciding what to do.” — Boris Cherny, Claude Code’s creator at Anthropic. AI tool builders started saying things like this in 2026. Instead of typing an instruction every single time, they design a system that runs on its own until the goal is actually done. This way of working is called loop engineering, and it’s spreading fast. This course teaches that design method, hands-on. It doesn’t stop at explaining the concept — by the end, you’ll have one real loop from your own work running automatically, every day.
The first half teaches six design components:
- How to hand off a goal instead of writing a step-by-step instruction sheet
- How to write a done-condition that’s true or false just by looking at the result
- Self-verification — having the AI grade its own output before it reports back
- Stop conditions, so a loop doesn’t run forever on a bad path
- Audit reporting — two sentences that remove the ability to fake progress
- Human approval gates, so nothing risky ships without a person clicking “go”






