Find bugs with property-testing agent swarms
Agent swarms doing property testing found about 30 bugs in the author's work code, then ~100 bugs across open-source projects like uv, with many fixes merged.
Done withCodex
- What they did
- The author gave a side-project talk at work on property testing with agent swarms. They had found about 30 bugs in their employer's code. To write a blog post that didn't center those, they pointed the same approach at open-source projects, including uv, codex and prometheus. The post doesn't describe the mechanics.
- How it went
- The author reports about 100 bugs found across the open-source projects, and says many fixes are already merged. The post gives no breakdown per project.
- Worth knowing
- The post is a brief announcement. It gives no cost, setup or tooling details, so look for the author's promised blog post for the method.
Try it yourself with Codex
Look at [my repo or module] and write property-based tests with [Hypothesis / fast-check / the framework for my language] for its core functions, such as parsers, serializers and data transforms. Run them, and for every failing case, shrink it to a minimal reproduction and tell me whether it looks like a real bug. Stop when you hand me a list of confirmed bugs with repro steps and suggested fixes, and do not change my source code without asking.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Bluesky · Oct 14, 2026
Related use cases
Turn tickets into merged pull requests
Pulls tasks from GitHub Issues or Linear, then runs Claude Code or Codex in isolated Kubernetes pods to carry each one through to a PR.
Rank CRM prospects and draft outreach with Codex
A demo has Codex pull CRM data, rank top accounts, and build a 13-slide deck with per-account strategy and draft outreach emails.
Cancel an Amazon subscription and get a refund
Codex navigated Amazon settings to cancel a subscription, then chatted with customer service and secured an immediate $15 refund.
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Vibe-code and deploy an idle clicker game
Gilbert Pellegrom had his OpenClaw agent build a simple idle game from a conversation, commit it to GitHub and auto-deploy it to Vercel.
Turn Slack feedback into PRs and debug infra
Sid Nanda's company uses Devin in Slack to triage feedback, turn user issues into PRs, debug async flows and infra read-only, and let non-technical staff coauthor PRs.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.