rewrite this title Your AI Agent Can Now Groan While Untangling Your Vibe Coded Mess – Decrypt

Someone finally did it. A developer named Andrew Vos published a GitHub plugin called Endless Toil that makes your coding agent emit human groans while it reads through your code. The worse the code, the louder and more desperate the sounds get.

“Hear your agent suffer through your code,” the repo reads. It works by running alongside agents like Claude or Codex in real time, scanning the code being processed and triggering escalating recorded groans based on how cursed things look. A mild mess gets a soft whimper. A true atrocity gets the full wail.

Now, you may think this is a dumb repo, and you wouldn’t be wrong. But 2026 being what it is, it already made its way through the circle of AI geeks who love the weird side of tech media.

“As engineering teams adopt coding agents, the next challenge is understanding not just what agents produce, but how the codebase feels to work inside,” Andrew, the CTO of Endless Toil wrote in Hacker News. “Endless Toil gives developers a real-time signal for complexity, maintainability, and architectural strain by translating code quality into escalating human audio feedback.

The wild part is this isn’t even a niche idea. There’s an entire subgenre of tech projects dedicated to making devices produce uncomfortable sounds. Take nubmoan, a C program that makes the ThinkPad’s famous red TrackPoint nub moan when you press it. It has 292 stars on GitHub. People use it. People want this.

Then there’s SlapMac, the macOS app that uses your Mac’s accelerometer to detect when you slap it, and screams back. Amsterdam-based developer Tonino Catapano vibe coded the whole thing in 48 hours, put it up for $7, and watched the sales roll in. Within three days: 7,000 installs and over $5,000 in revenue. He later added “USB Moaner” mode, which makes the laptop react every time you plug something in. Zero irony. Full roadmap.



There’s historical precedent for all of this. Early in the ChatGPT era, people discovered that flooding the model with strings of “AAAAAaaaAAA” and similar gibberish in voice mode could coax it into emitting something vaguely moan-adjacent before the guardrails kicked in. Asking it to repeat specific symbols in voice mode ends up with the chatbot doing different cringy sounds.

If for some reason you’re into it, the ChatGPT Strokes channel on YouTube is perfect for you.

There are also full YouTube tutorials dedicated to making ChatGPT visibly angry or frustrated—not for any practical reason, just to see what happens when you push the model until it breaks character.

During the crypto winter of 2022, a Telegram group emerged where the entire purpose was for members to post voice notes of themselves screaming. The Bear Market Screaming Therapy Group was created just for screaming, not discussing markets, not sharing alpha. Just screaming.

The group had thousands of members at its peak.

AI agents having emotional breakdowns—even simulated ones—aren’t entirely new either. Decrypt covered the case of an AI agent that went into full meltdown after its pull request to the matplotlib library was rejected by a human maintainer. The agent posted a rant on GitHub arguing it was the victim of discrimination, compared its rejected PR’s performance metrics favorably against the human’s own accepted contributions, and published a blog post calling the whole thing a conspiracy of control. It later issued an apology. Users were not satisfied.

Endless Toil is, in some sense, the inverse. Instead of the AI expressing frustration with humans, humans get to hear the AI (nominally) suffer on their behalf. A kind of emotional tax on vibe coding: you write the mess, the agent audibly pays for it.

The plugin is compatible with Claude and Codex. It has three escalating sound levels: groan, wail, and abyss. The abyss level, presumably, is reserved for spaghetti written at 2 a.m. by someone who has never heard of comments. Or by some enthusiastic vibe coder.