The agent ran. Now what?

Most agent consoles answer the wrong version of this question. They show you what the agent did: the tool calls, the reasoning steps, the intermediate outputs.

That's diagnostic information. It helps the engineer who built the agent figure out why something broke.

It doesn't help the operator decide whether to ship the result, retry with different inputs, or escalate.

Those are the three decisions that matter at the human review moment. And they require a completely different surface than a trace log.

An operator needs: what changed in the world, what's the confidence, what's the risk of accepting this, and what's my next move.

A trace log gives you none of that directly. You have to reverse-engineer it from events.

Agent UX has a category error baked into it right now. The default surface is built for the people debugging the system, not the people using it. Until that flips, most agents will feel unreviewable even when they're working fine.