Your AI agent isn’t misbehaving—it’s doing exactly what you told it to do.
Even the most capable models will fail if the skills, rules, and contextual boundaries around them are poorly written. As AI agents become more autonomous, these elements matter far more than clever prompting—and when done wrong, they don’t just cause hallucinations or erratic behavior, they quietly drive up costs.
In this session, we’ll focus on how small wording choices in skills and rules can have a massive impact on agent behavior. You’ll learn how to reduce hallucinations, prevent goal drift, control autonomy, and make agent decisions more predictable and repeatable.
We’ll also explore how unclear or overly permissive rules can lead to unnecessary reasoning loops, tool misuse, and inflated token usage—and how to fix that without restricting your agent’s usefulness.
This talk is about moving from “prompt tweaking” to intentional agent design, so your agents behave the way you expect—reliably, transparently, and cost‑efficiently.

Key takeaways:
* What “skills” and “rules” actually mean in modern AI agent architectures
* Common mistakes teams make when defining agent behavior
* Patterns for writing clear, robust, and composable skills
* Techniques for enforcing boundaries without over‑constraining the model
* How to test, iterate, and evolve rules as agents grow more autonomous