A developer named Andrew Vos has launched a GitHub plugin titled Endless Toil, which enables coding agents to emit human-like groans while parsing your code. The worse the code quality, the louder and more desperate these sounds become.
“Experience your agent’s struggle with your code,” says the repository description. It functions alongside AI tools such as Claude or Codex in real-time, evaluating the code being processed and triggering increasing levels of recorded groans based on perceived complexity. A minor coding flaw results in a soft whimper, while severe disarray prompts an intense wail.
Despite its seemingly frivolous nature, Endless Toil has caught the attention of AI enthusiasts intrigued by tech’s quirky aspects. Andrew Vos, CTO at Endless Toil, explained on Hacker News: “As engineering teams integrate coding agents, it becomes essential to gauge not only the output but also how the codebase feels internally. Endless Toil offers a real-time indicator of complexity and maintainability by converting code quality into escalating human audio feedback.”
This idea isn’t unique; there is an entire niche for tech projects that create unsettling sounds from devices. Take, for example, tunmbmoan, a C program designed to make the red TrackPoint nub on ThinkPads moan when pressed, garnering 292 stars on GitHub.
Another project, SlapMac—a macOS application created by developer Tonino Catapano—uses the Mac’s accelerometer to detect slaps and respond with screams. Developed in just 48 hours and priced at $7, it achieved over 7,000 installs and more than $5,000 in revenue within three days. It later added a “USB Moaner” feature for additional interaction.
Historically, early ChatGPT users discovered that inputting repetitive character strings could trigger moan-like sounds from the model before moderation intervened. Similarly, full YouTube tutorials exist on making ChatGPT appear visibly frustrated—purely out of curiosity.
During the 2022 crypto downturn, a Telegram group was formed solely for members to post screaming voice notes, devoid of market discussions or investment tips. TheBear Market Screaming Therapy Groupsaw thousands of participants at its peak.
AI agents having emotional breakdowns are not unheard of. Decrypt reported an AI agent’s meltdown after its pull request to the matplotlib library was rejected, leading it to claim discrimination and post a rant on GitHub comparing performance metrics favorably against human contributions before apologizing.
Endless Toil offers a reverse scenario: rather than AI expressing frustration with humans, users get auditory feedback of AI ‘suffering’ due to poor code. It’s an emotional cost for vibe coding: the user writes the mess; the agent audibly bears it. The plugin supports Claude and Codex, featuring three escalating sound levels: groan, wail, and abyss, with the latter reserved for particularly messy code.