Why is intake a good test for AI agents?

**Why use intake to test agents rather than a chat task?** Because the agent must change state in connected systems. It reads a submission, looks up a customer record, checks documents against policy and writes a routing decision. Each step can be checked against a recorded expected result.

**Isn't intake too simple?** The happy path is simple. The difficulty sits in exceptions: duplicates, partial submissions, conflicting fields, expired documents, and cases that match two routing rules. A task pack can weight these heavily so an agent that only handles clean forms is visible quickly.

**What does a wrong answer cost in a real business?** Mis-routed cases sit in the wrong queue, incomplete files reach specialists, and duplicate records split a customer's history. Those costs are why the workflow suits evaluation: errors are concrete and attributable to a step.

**Does the data come from real applications?** No. The material is generated. It is informed by direct observation of how real businesses handle intake, but a release does not contain real applicant records. Buyers should still inspect generated samples for grounding and for any resemblance to real people or organisations.

**Where should a team start?** Pick one intake path, such as insurance claim lodgement or utility connection requests, and ask for a small task set covering clean, incomplete and duplicate submissions.