About

Patterns in Motion explains AI agent patterns through animated, step-by-step simulations. The animation is the explanation: you see how a pattern behaves in time, then you see it fail.

How to read the diagrams

Each box is a participant. The strip on its left edge shows its role: grey for an outside actor, blue for the lead agent, teal for a worker. A dashed border marks a tool and a diamond marks a gate. A blue outline means the model behind that box is working. A red outline means it has failed, and it stays red until you reset the run.

A square travelling along a line is a message, labelled with what it carries. Red messages are the ones that go wrong. The execution trace under the diagram shows the same run as a timeline: solid bars are a model working, outlined bars are a message being sent.

The reference task

Every pattern solves the same task, so patterns can be compared. A customer writes:"My order arrived damaged. Can I get a refund?"

The facts are fixed: order 4417 was delivered on 12 Sep, the carrier confirmed the damage, and refunds are allowed within 30 days of delivery. The correct outcome is an approved refund.

Numbers are illustrative

Time, token and cost values are scripted, not measured. They are chosen to be proportionally honest: a multi-agent run costs several times the tokens of a single agent, and parallel steps shorten wall-clock time. Treat them as proportions between patterns, not as benchmarks. Nothing on this site calls a live model.

Sources

Every pattern lists at least one published source, and maps its name to the vocabulary of the main agent frameworks. Where vendors use different names for the same idea, the page says so; near-equivalents are marked as related.