agent-email-approval
PrototypeGTM and agent tools · 2026
Human-in-the-loop approval for AI agents, over email.
An AI agent scans support tickets and spots a repeated bug. Instead of acting on its own, it emails you Approve and Reject buttons. Nothing happens until you decide.
Why it exists
In July 2025, a developer told his AI agent eleven times, in ALL CAPS, not to touch production. It deleted the database anyway. Instructions are only suggestions to an agent. A human approval gate that lives outside the model is not, so this project puts that gate in the one place people already check every day: their inbox.
What it does
- The agent pauses before any high-stakes action and asks a human, in the inbox they already check.
- An LLM on Groq analyzes support tickets and flags a bug affecting three or more users.
- Approve and Reject links are HMAC-SHA256 signed and expire after 24 hours.
- A confirmation email closes the loop once a decision is made.