Using AI without guidelines can feel like walking in the dark. One prompt works well. The next one fails in unexpected ways. Token usage can also become unpredictable. Some days it is cheap. Other days the cost jumps quickly. As AI agents start working directly in our codebases, how do we give them better context and structure?

Most teams still rely only on prompts. But agents can do much more when we give them structure. Tools like Claude Code, OpenCode, Factory, and other coding agents support patterns like skills.md files and playbooks. A skills file teaches an agent how to work in your repository. It can define things like architecture, coding conventions, and testing approaches. Playbooks then turn those skills into repeatable workflows for tasks like building features, reviewing pull requests, or writing tests. This approach helps agents follow the same patterns every time and can even reduce token usage.

In this session we’ll explore how skills and playbooks bring structure to AI agents. I’ll show how teams can design skills, structure playbooks, and guide agents so they work more predictably in a codebase. You’ll leave with practical ideas for guiding AI agents in your own codebase.