Skip to content

eshagarwal/Human-in-a-loop

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 

Repository files navigation

Human-in-a-loop

Tap to Trust — Designing the Approval Moment for Agentic AI

🔗 View Prototype

Designing the 10-second decision moment between an AI agent asking for approval and a business owner confidently tapping back.

Overview

As AI agents evolve from simply chatting to actively taking actions for businesses, the challenge is no longer just automation — it’s trust.

This project explores how SMB owners can stay in control of high-stakes AI actions without slowing down their workflow.

Instead of limiting the experience to a simple Approve / Deny, we designed a mobile-first approval system that supports:

  • quick approvals,
  • conditions,
  • modifications,
  • AI guidance,
  • takeovers,
  • deferrals,
  • and learning loops.

Scenario

The Owner

Arjun Mehta — Manager of The Willow Boutique Hotel, handling guests mid-shift.

The Customer

Maria Gonzalez — Returning guest requesting a booking change and room upgrade through WhatsApp.

The Agent

Lily — The hotel’s AI assistant.

Lily checks availability and pricing, but since the booking upgrade exceeds the auto-approval threshold, the request is escalated to Arjun.


The Approval Moment

Arjun receives a lock screen notification:

🏨 Lily needs your call
Maria G. · Booking change + upgrade · +€255 · June 7–10

He taps and sees a fast, glanceable approval interface.

Maria Gonzalez                     ★ Returning Guest (3rd stay)
────────────────────────────────────────────────────

CURRENT                 →                 REQUESTED

June 3–5                                  June 7–10
Standard Room                             Garden Suite
€180 total                                €435 total

────────────────────────────────────────────────────
                    +€255
                [ 🟢 Low Risk ]
────────────────────────────────────────────────────

What would you like Lily to do?

Key Design Principles

Context, Not Clutter

The owner sees only what matters:

  • action,
  • customer history,
  • pricing impact,
  • and risk level.

Additional details are revealed progressively only when needed.


More Than Approve / Deny

The system supports a full spectrum of responses:

  • ✅ Quick Approve
  • 🔢 Approve with Conditions
  • 💬 Whisper to the AI
  • ✏️ Modify Request
  • 👋 Take Over Personally
  • ⏰ Defer
  • ❌ Deny with Reason

This creates collaboration between human judgment and AI automation.


Customer Experience During Waiting

While the owner decides, Lily keeps the customer informed naturally:

“Thanks for your patience — I’m just confirming a few details for you.”

This prevents silence from feeling like failure.


Adaptive Learning

After repeated approvals, Lily learns preferences over time:

“Should I auto-approve future upgrade requests under €300 for returning guests?”

The AI gradually becomes more personalized and trustworthy.


What We Focused On

  • Mobile-first interaction design
  • Notification UX
  • Progressive disclosure
  • Emotional trust
  • Human-AI collaboration
  • Failure handling
  • Adaptive automation

Final Thought

The future of AI products is not just about automation.

It’s about designing moments where humans and AI can make decisions together — quickly, confidently, and naturally.

License

This project is licensed under the MIT License — see the LICENSE file for details.

About

A human-in-the-loop mobile-first approval system for AI agents handling high-stakes business actions.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors