Testing AI Agents in TypeScript Without Calling the Model

Most agent bugs are in your loop, not the model. A scripted fake client that makes runaway loops, tool errors and budget stops deterministic unit tests.