You ask Claude Code to fix a broken function. It confidently spits back the exact same broken code, just with different variable names. You stare at the screen, your blood pressure spikes, and you type: “No, you f***ing idiot, I said fix the logic.”
You aren’t losing your mind. You aren’t unprofessional. You are experiencing a textbook trust failure.
Recently, a developer named Marcel Petrick got so frustrated testing AI agents that he created a GitHub repository called AgenticSwearJar to track exactly how often he cursed at Codex and Claude Code. He guessed maybe 10% of his prompts contained profanity. What he actually found was a fascinating window into human-machine psychology.
Profanity isn’t a bug in the human-machine interface; it’s the highest-signal telemetry we have.
Think about it. You don’t swear at a toaster when it burns your bread. You just push the lever again. But when an AI agent fails, we curse. Why? Because these models have crossed a threshold. They converse with us, they reason with us, and they apologize to us. They feel human enough to trigger our social expectations—and when they fail, they trigger our social rage.
Yet, at the exact same time, we expect flawless, deterministic perfection from them. We are blaming the AI for being too human while judging it as a non-human system that should never make mistakes.
We curse at machines because they feel human enough to frustrate us, yet we judge them as flawless deterministic systems.
This tension is exactly why your profanity is so incredibly valuable. Right now, developers and AI companies are scrubbing swear words from prompt logs to keep things looking clean and professional. This is a massive mistake.
When you drop an f-bomb in a prompt, you aren’t just venting. You are leaving a massive, glowing breadcrumb trail pointing directly to the exact moment the AI violated your model of its expected capability. You thought it could do X. It did Y. The trust broke. You swore.
Instead of being filtered out, swear words should be treated as UX friction flags. If a thousand users start their prompt with “Why the hell did you…”, that isn’t noise. That is an AI agent’s failure mode flashing in neon lights.
Stop filtering the profanity out of your prompt logs. A swear word is just a UX friction flag with a pulse.
So the next time you’re yelling at Codex for breaking your build, don’t feel bad about your language. The AI should be logging that anger, tracing it back to the exact step where it lost the plot, and learning from it. Your frustration isn’t a flaw in your character. It’s the most honest feedback the machine will ever get.
FAQ
Q: Isn't swearing at an AI just a sign of poor impulse control?
A: No, it's a predictable psychological reaction to a system violating your expectations. When an AI acts human but fails at a basic task, it triggers a social response. That anger is data.
Q: How can AI companies actually use profanity as telemetry?
A: By flagging prompts with high emotional intensity or profanity to identify exactly where models consistently fail user expectations. These friction points can then be targeted for retraining or better guardrails.
Q: Should we actually encourage developers to swear at their AI?
A: We shouldn't encourage the frustration, but we absolutely shouldn't penalize or filter it. Let the user vent naturally, and let the system learn exactly where and why it broke their trust.