Skills Teach AI How
A model knows an astonishing amount about everything and nothing about how you do anything. How you like reports structured. What your review checklist actually checks. Which steps your process never skips. Ask a bare model to do your task and it does a version — competent, generic, and not yours. Sound familiar? That gap has a name, and a surprisingly simple fix.
A skill is your method, written down, that the AI reads before doing the task. Think of a page in an employee handbook, or a recipe: the steps, the standards, the things to never do. When the work comes up, the AI pulls the relevant skill and follows it — your way, every time, instead of reinventing a generic way each time.
Three properties make skills more than documentation. They’re executable — the AI doesn’t just store them, it acts on them. They’re composable — small skills combine into bigger behaviors, the way good notes link into bigger ideas. And they’re improvable — when the output misses, you don’t retrain anything; you edit a file, and the next run is better. Teaching an AI starts to feel less like programming and more like training a new hire who never forgets.
The industry has converged on this pattern hard — every major AI platform now ships some version of skills as a first-class feature. The know-how layer is becoming as standard as the connection layer.
Related
- Personas: A Specialist You Shaped — what skills combine into.
- The Brain Is the Structure Around the Model — skills are part of the structure you own.