Skills as the Interface to Agentic Products
- Source
- youtube.com
- Author
- AI Engineer
- Date
Context
An can package a product workflow as instructions the model loads when it recognizes the user's intent. This talk describes a registry containing each skill's name, description and file location. The model sees those summaries first, then reads the selected instructions instead of receiving the entire library upfront.
That makes the description part of the product's routing behavior. A request for a PDF report must lead to the PDF workflow rather than another report format. The speaker also recounts a model upgrade that changed how instructions were followed even though the skill file stayed unchanged. Testing the model and skill together checks whether the intended workflow still happens.
- agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
- embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
“Skills are the new features.”
Yogi
“Skills without evals are really just wishful thinking.”
Yogi
“Skills are not the documentation and a lot of people treat them like that and skills are really the contracts versioned to a model.”
Yogi
videoHow to Write Your First AI Eval - SallyAnn DeLucia, Arize AI
videoBuild Your Own Team of Managed Agents - Jack Wotherspoon, Google DeepMind
videoHow to Build an AI Code Review Agent - Daksh Gupta, Greptile
videoHow to Build Voice Agents That Don't Fall Apart When You Interrupt - Jesse Hall, LiveKit
Checking sign-in…
Loading comments…