Tap to Trust

Designing the Approval Moment for Agentic AI — a Superchat design challenge.

The 10 seconds between an AI agent asking for the go-ahead and the owner tapping back. Not just Approve/Deny, but the full spectrum: constrain, modify, whisper, takeover, deny-with-reason, defer.

Prototype

Interactive mock-up built with Lovable → roger704/hands-free-help

Presentation

Slidev deck that pulls the live Lovable mock via iframe — see presentation/.

cd presentation && npm install && npm run dev

Contents

File	What it is
Challenge.md	The original challenge brief, structured.
scenarios.md	10 end-to-end scenarios (dental, restaurant, PT) — each with the full 6-step flow.
scenarios.excalidraw	Those scenarios as a 3×3+ grid of cards.
tap-to-trust-flow.excalidraw	Full briefing board — problem, context, personas, focus areas, flow, response spectrum.
figjam-board.svg	SVG of the briefing board — drag into FigJam.
figjam-flow.mermaid	Flow + response spectrum as Mermaid — paste into FigJam's Shift+D diagram tool.
figjam-paste.txt	Plain-text version — paste into FigJam, becomes stickies.
sample-data.json	Realistic approval-request fixture data for prototyping.

The 6-step flow every scenario follows

1. Trigger — customer message or system event
2. Agent detects — recommendation + confidence + risk signals
3. Owner sees — the glanceable notification
4. Owner decides — one tap on the response spectrum
5. Agent resumes — in-character, without breaking the conversation
6. Outcome — result + learning signal