Talk
You’re Writing Too Many Skills!
Whenever an AI agent fails at a task, our immediate reflex is to write a new skill for it. Tutorials, team coding standards, one-off prompts—suddenly everything is packaged as a skill. But when we treat skills as a dumping ground for team knowledge, we end up with bloated contexts, confused agents, and duplicate documentation that quickly goes out of date.
Just because something can be a skill doesn’t mean it should be.
Join this talk to discover why building your team's documentation on an AI framework that could be deprecated in a week is a bad idea, why many of the skills teams write probably shouldn't exist, and how to spot the warning signs in your own repositories. We’ll cover what a good skill actually looks like—and what you should be doing instead to make your agents genuinely more reliable.
About Martyna Wybraniec
Martyna Wybraniec is a Software Engineer at Google based in Warsaw, working on Google Cloud Networking. Passionate about developer productivity and tooling, she actively explores the intersection of software engineering and agentic AI—focusing on how to build reliable, maintainable workflows around AI coding assistants.