Appearance
Day 3: Agent Skills
An agent that forgets everything the moment a conversation ends is stuck relearning your business, your codebase, and your preferences from scratch every single time. Today is about fixing that, giving an agent memory that survives past one session, teaching it to manage a long context window without burning your budget, and showing it how to write down what it's learned as a reusable skill instead of rediscovering the same solution over and over. By the end of today, "skill" stops being a vague capability and becomes a specific artifact your agent can build and reuse.
What you'll cover
- Giving Agents Memory , storing and retrieving your own data so an agent's answers are grounded in something real.
- Long Context & Token Economics , managing context window size and cost as conversations and tasks grow.
- Building Reusable Skills , how an agent can write and reuse its own procedural skills, using the real mechanism a production framework ships.
- Multi-Agent Workflows & State , coordinating more than one agent on a task, with checkpoints and a human-approval gate for anything consequential.