Relay — Failure Recovery & Verification for AI Coding Agents

Keeps AI coding agents working through failed tests, missing deps, and small mistakes

Problem

Many developers find that AI coding agents are only good for about 30 seconds before they completely fail on the slightest hiccup: a failing test, a temporarily unavailable dependency, or an incorrect assumption. Agents are often good at doing something slightly wrong, but not at recovering from it, so humans end up babysitting them anyway.

Solution

A lightweight harness that runs alongside AI coding agents, continuously monitors their output and test results, automatically rolls back to the last stable state when a small failure occurs, corrects wrong assumptions, and re-executes. It also generates a verification report so users know exactly what passed and what didn't.

Users

Individual developers and small teams who heavily use AI coding agents like Claude Code, Cursor, or Aider.

MVP

CLI that integrates with major coding agents; automatic rollback to the last successful snapshot on failure; built-in recovery strategies for common errors (failing tests, missing dependencies); readable execution logs and verification summaries; simple notifications (macOS notifications or Slack webhook).

Revenue

Open-core with a paid cloud sync/team collaboration tier; or subscription based on agent runtime or number of runs.

Why now

AI coding agents are rapidly being adopted, but reliability remains a major pain point. An indie developer can build a local CLI tool without heavy infrastructure and iterate quickly.