"They've gotten me sick of LLM writing in a hurry" — after hours of reading AI output every day, the prose itself started to feel toxic

Alec Scollon spends hours every day reading AI-generated text. His job involves building frameworks for large-scale unsupervised code generation — when he's not pairing with Claude on tooling, he's sifting through Qwen agent output. For casual questions, he reaches for ChatGPT or Gemini over search, especially when AI-generated articles clutter the results.

His work has shifted dramatically. What used to be designing and writing code is now a four-step process: designing code, describing the design to an LLM, reviewing what the LLM produces, and finally writing code when needed. He feels more productive and more comfortable in unfamiliar domains. But in the last few months, a small part of him started to dread reading LLM output — because he already knows what he'll find.

False assumptions and hallucinations. Emphatic, staccato fragments. Excessive emojis. The same rhetorical patterns across every model.

"On their own, none of these annoyances gets to me," he wrote. "Together, though, they've gotten me sick of LLM writing in a hurry."

He is explicit that he is not condemning LLMs — humans are unreliable too. The problem is repetition: models write in the same style and make the same classes of mistakes. Personalization settings help, but idiosyncrasies seep through, and he can't control prose from other people's agents.

His post on Hacker News gained 154 points and 101 comments, with many developers echoing the same experience. Commenters called it "clanker fatigue," "assembly-line QA," and in one case, reported feeling physically ill after hours of Opus output. The burnout is cumulative: supervised review plus unsupervised slop plus AI-shaped search summaries.

Scollon's methods are what he calls "probably primitive" — one task at a time, reading output thoroughly before revising. He's not deep in agent orchestration. Yet the exhaustion has already set in. The question he's raising isn't whether AI is useful — it's what happens to human attention and craft when the default mode of work becomes reviewing machine-generated text.