Everyone tunes their prompts, switches models, reaches for bigger context windows, and Claude keeps floundering in the code. Why? Maybe the fix isn't the one we think it is. In this talk I'll walk you through what I've learned in the trenches and what I had to unlearn.

For developers using coding agents and technical leads shaping their adoption. The hook is a real paradox: the same agent behaves differently depending on the code it reads. The session gives the audience concrete ways to examine consistency, locality and readability, plus a starting point for improving a legacy codebase. Stack-independent, with a story already presented at Claude Torino. Available as a lightning talk or a regular session.
Recording in Italian: https://www.youtube.com/live/YBxcoCuHvpY