Agents That Debug Humans

Even the smartest AI systems can fail for very human reasons - like a user misconfiguring an instruction or overloading an AI “railway.”

In this live coding demo, I’ll show how we built agents that debug humans: detect and repair faulty user configurations before they ripple through production.

You’ll see how just a few lines of code can help agents interpret intent, rewrite problematic rules, and verify improvements automatically - the same principle that powers large-scale AI organization at NEXT.

This talk explores how we’re designing agents that make AI systems resilient to human error - turning configuration chaos into stable, intelligent workflows.