The product problem
Most agent demos optimize for outcome fluency. Operators care about trust, repeatability, and proof.
- What page did the system choose?
- What element or field did it actually trust?
- What should a human check before acting on the result?