What does it mean that agents operate on schemas?

Every agent embeds assumptions about the world — the schema it uses must be accurate.

Example: You built an agent for handling criticism at work: 'When someone challenges my idea in a meeting, assume they are attacking my competence, and defend the idea immediately.' This agent fires reliably. It also destroys collaborative relationships. The schema it operates on — 'challenges equal attacks' — is wrong. Replace the schema with 'challenges are stress tests for the idea, not evaluations of the person,' and the same trigger produces a completely different action: listen, ask a clarifying question, improve the idea.

Try this: Pick one agent you already run — a repeatable behavior triggered by a specific situation. Write down the schema it operates on: what does this agent assume about the world? Then ask three questions. First, where did this assumption come from? Second, when was the last time I tested it? Third, what would change if this assumption were wrong? If you cannot answer the third question, the schema is running unchecked.