All studies01 / 74
AI interface
Agent approval
The card an agent raises before it acts: the chosen option grows into the outcome and the alternatives collapse away.
Where it earns its place
Human-in-the-loop gates in AI products: publishing, sending, deleting or spending, anywhere a person must own the decision.
What we changed
- Set on a real gate: the agent asks before publishing the Northstar Cloud launch page, with the preview facts it checked.
- The chosen button itself becomes the outcome (shared layout), instead of being swapped for a separate resolved row.
- Gave each option its own end state (shipped, on hold, discarded) and made the destructive choice merlot rather than a generic red.
- The demo picks an option on a loop but pauses as soon as a pointer or keyboard focus is in the card, so visitors can decide themselves.