Every case should answer five questions before anyone trusts an AI output:

  1. What is the user trying to get done?
  2. Which workflow owns it?
  3. Which current sources govern it?
  4. What can the assistant draft or recommend?
  5. What still requires a human decision or real-world action?

<aside> 💡

Open a populated case below and follow it through the whole loop. Use START HERE — duplicate this case when you want to try your own input.

</aside>

Three useful paths

A recommendation is not an approval. A draft is not sent. A handoff is not proof that work moved.

Untitled

Untitled