A Developer's Case of 'LLM Burnout': Sick of the Same Hallucinations and Emojis

A software developer who leans on Claude Code and Codex for hours daily describes a creeping fatigue with LLM output. His workflow has inverted over the past year — where he once designed and wrote code directly, he now designs, describes the intent to an assistant, reviews what it generates, and only then writes code himself. He credits the tools with exposing him to unfamiliar approaches and making him more comfortable in areas outside his expertise, and his current project is building tooling for large-scale, unsupervised code generation, which means he spends most of his day reading machine-written text either way.

The burnout isn’t about any single flaw but their relentless repetition. He’s grown to dread reading model output because he can predict what he’ll find: confident false assumptions, hallucinations, clipped staccato phrasing, and a scattering of emojis. None of these bothers him in isolation, but LLMs share a house style and make the same categories of mistakes over and over, and personalization settings only paper over some of the idiosyncrasies. He can’t control the style of AI text produced by other people at all.

Notably, he isn’t condemning the technology — he expects to keep using it, still finds it a net productivity gain, and values learning to wield it well. His point is more unsettling: it’s not just frustration at a flaky tool, but a genuine aversion to the aesthetic and cognitive sameness of machine-generated prose, a fatigue he didn’t anticipate and doesn’t yet know how to manage.