Beautifully Wrong Entry 005 of 034

Future Defaults · No. 005

The handoff console

When agents do the work, the interface's real job changes: not buttons for doing, but instruments for trusting. This console shows what each agent is doing, where it is uncertain — and exactly one moment that needs a human.

Working for you

3 agents · 1 needs you

since 09:12
Travel — rebooking Lisbon running

Comparing 14 routes. Holding two refundable fares.

confidence 86%
Email — drafting the apology to Verde & Co. needs you

I can absorb the blame or explain the vendor delay. These send very different messages. Which is true?

confidence 41% — paused
Research — Q3 competitor scan drifting

Sources thinning. Will surface a partial brief rather than pad it.

confidence 63%
Everything else is safe to ignore. Next check-in: 11:30.
Fig. 1 Interactive — make the one decision the agent paused for.

What it breaks. The dashboard. Status pages monitor machines; this monitors delegation. It also breaks the chat transcript — you shouldn't have to scroll a conversation to find the one question that matters.

Why it might work. Trust in agents will be built or lost at the handoff points. An interface that volunteers its own uncertainty — and pauses exactly there — treats attention as the scarce resource and judgment as the human job.

Where it fails. Confidence percentages are seductive fictions. An agent that is wrong with 86% confidence is more dangerous than one that asks too often — and a calm console can launder that wrongness into trust.

If the agent only interrupts you once a day — how good does that one interruption have to be?