I think I have LLM burnout
Programmers describe feeling exhausted and demoralized by the shift from writing code themselves to supervising large language models that churn out vast amounts of often low-quality “slop” code and text. While some enjoy the productivity boost and new possibilities for side projects, many report constant context-switching, pressure to meet AI-inflated expectations, and the cognitive strain of endlessly reviewing and correcting machine output. Commenters liken this to earlier waves of automation that turned skilled craft into repetitive oversight work, and predict more developers may leave the field as the job becomes less about creating and more about managing opaque, error-prone systems.
Nature of LLM burnout
- Many describe a shift from “building things” to “design → prompt → review → babysit,” which feels like managing an unreliable junior or “bullshit artist” at scale.
- Burnout comes less from using LLMs personally and more from reviewing endless AI‑generated code, docs, and plans from others.
- People report cognitive fatigue, zoning out, even mild physical symptoms (dizziness, nausea, “psychic damage”) from reading so much similar LLM prose.
Quality, testing, and verification burden
- Core complaint: generation is cheap; verification is not. Review and QA become the main bottlenecks.
- Some lean heavily on exhaustive tests, strong typing, and end‑to‑end checks to constrain agents, but note that AI‑generated tests can also be wrong, vacuous, or fragile.
- Many say they can’t reliably review LLM code faster than they can write it, unless they accept sloppier quality.
Team dynamics and “slop”
- Common pattern: weak or non‑programmers use LLMs to produce large volumes of barely‑understood code or docs, then offload review to others.
- This “slop at scale” affects senior engineers most, who feel responsible for preventing the codebase or documentation from degrading.
- Some organizations reportedly mandate LLM use and even track token usage, creating perverse incentives to generate more artifacts than can be responsibly checked.
Style fatigue and “botspeak”
- Strong aversion to LLM default tone: hypey, repetitive phrases, emojis, clichés, overuse of certain words, and dense jargon.
- People note that models develop recognizable idiolects; reading them all day feels like “fast food language.”
- Some mitigate this with strict style guides (no emojis, banned phrases, specific voices), but hallucinations and shallow reasoning remain.
Productivity, expectations, and addiction
- Individual output can jump 10–20×, enabling solo devs to ship ambitious projects and many side projects.
- That same boost drives pressure: always “one more task,” always another agent to spin up, difficulty stopping work.
- Several compare this to the industrial revolution or assembly lines: more throughput, but more monotony and higher expectations.
Career and identity
- A substantial contingent says LLM‑centric work has made them question programming as a career or abandon coding as a hobby.
- Those who love the process and craft of programming often feel especially alienated; those who are more product‑focused tend to embrace the tools.