LizardBuild: 50 parallel Claude Code sandboxes race to fix one bug
LizardBuild snapshots a buggy sandbox into 50 forks, each running a Claude Code agent with a different idea; the first green test wins, the rest pause, billed per second.
50 sandboxes. 50 claude code agents. 1 bug. i paid for exactly one of them the other 49 sat on pause at $0 while the winner shipped the fix here's the setup on @LizardBuild: → one sandbox with the bug, snapshotted → 50 forks with memory and running processes, each ready in 1–3 seconds → every agent gets the same bug and a different idea → first green test wins, the rest pause instantly → billed per second. 100 small sandboxes = $1.62 an hour broken deploy at 3am? the error comes back as JSON with the reason and exit code. the agent reads it and fixes it before you wake up you stop debugging. you start picking winners
This post quotes another post: x.com/lizardbuild/status/2111803353986380052