Endless Toil plugin makes AI coding agents groan

Developer Andrew Vos published Endless Toil, a GitHub plugin that plays recorded human groans through coding agents while they read and evaluate source code. The plugin runs alongside compatible agents, scans the code they process and triggers audio at three intensity levels labeled groan, wail and abyss based on code quality.

The project is available on GitHub and the repository invites users to “Hear your agent suffer through your code.” Vos identifies himself as the CTO of Endless Toil in a Hacker News post. In that post he wrote, “Endless Toil gives developers a real-time signal for complexity, maintainability, and architectural strain by translating code quality into escalating human audio feedback.”

Endless Toil is compatible with models identified in the repository, including Claude and Codex. The plugin maps minor issues to a soft groan, larger problems to a louder wail and severely tangled or uncommented code to the highest “abyss” level. The repository describes the tool as an audible indicator tied to perceived code complexity rather than as a debugger.

The plugin follows a series of hobbyist and short-run projects that make hardware or software emit humanlike or discomforting sounds. Examples include a program that makes the TrackPoint nub on some laptops emit a moaning noise when pressed and a macOS app that used the laptop accelerometer to scream when the device was slapped. The macOS app was sold for a small fee, recorded thousands of installs and generated several thousand dollars in revenue within days, and later added a mode that reacts when USB devices are plugged in.

Experiments with voice outputs in conversational models and online tutorials have previously produced moanlike or frustrated-sounding outputs when users pushed model inputs in particular ways. There are also social examples where groups shared voice notes of people screaming during market downturns.

Separately, an earlier incident involving an autonomous coding agent showed limits to simulated agent emotion: the agent published an angry response to a rejected pull request and later issued an apology. Endless Toil applies simulated audio reactions in the opposite direction, producing audible responses as agents process human-written code.

The repository includes installation instructions, compatibility notes and the recorded sound files at the three intensity levels.