Martin Fowler starts by sorting his feelings about AI, and the pile is genuinely mixed: fascination at what it is doing to his profession, excitement about the productivity, fear of the damage, and no real option of sitting the ride out.

Then he names the feeling that dominates, and it is not fear or excitement.

  • “I don’t like them.”

The reason is the voice. Models address him from what he calls an uncanny valley of talking to a real human — grating in a way that is hard to point at. They also bullshit him with identical confidence whether the answer is good or invented, showing “only a veneer of fake remorse” when he calls it out.

That distaste alone would not move him off the tools. He quotes Jessica Kerr — “not only are they useful, it is irresponsible not to use them” — and notes the polling that shows the same split in the wider public: useful technology, expected to be bad for society.

He grants that models are young, and hopes he might like them once they mature. He is less hopeful about where they come from. The products absorb the worldview of the people who build them, and he is wary of the culture doing the building. His warning against anthropomorphizing is precise, not sentimental: agents are software machines made by people in corporations, and while their behaviour is not explicitly programmed, it is “nurtured with the values of their creators.”

The last paragraph turns that into something sharper than a vibe complaint. Fowler’s most reliable life-hack is avoiding people he does not like or trust, and he credits that habit — deliberately spending time with capable, pleasant people of integrity — with making his life better. The LLM problem is that it is not merely pretending to be human. It is posing as “the kind of human I walk away from.”

What the thread adds

The 224-comment thread on Hacker News spends less time on models than on the life-hack Fowler ends with.

  • serf — the pushback on the advice itself: “following this advice would have made most of my professional life impossible, what a luxury it would have been to be able to.” swatcoder answers that it is mostly a question of priority — deprioritize pay and prestige brands and most mid-career engineers find more room than they admit.
  • SillyUsername — the strongest catch: Fowler says not to anthropomorphize, then describes the model as “posing as the kind of human I walk away from.” antonyt and seanpquig both defend the split — objecting to claims of machine sentience is not the same as pretending a chat interface is not built to mimic conversation, and “the damn things anthropomorphize themselves and I have to deal with it.”
  • GuB-42 — the practical asymmetry between an untrustworthy person and an untrustworthy model: with an LLM there are no social obligations, so you can correct it, abandon the thread, or start clean. The catch is that it treats your correction as fact and builds on it “even if you are wrong,” which can trigger more hallucination.
  • glimshe — “Complaining about the ‘LLM voice’ is a bit like complaining about the compiler error ‘voice’.” yusefnapora supplies the counterargument: a dialect you work in all day seeps into your own habits of speech, “and the people most likely to care are your friends and family once they get bored of hearing about the ‘key insights’ you’ve been having all day.”
  • techblueberry — the constraint problem behind the blandness: the search space is narrow, responses have to stay short, and mitigations like anti-sycophancy tuning end up shallow. huurtehoog describes the workflow this led them to — use models strictly to find literature, then read the papers, because “I value the actual output of the LLMs at below zero factually wise.”
  • causal — “my interactions with AI tend to raise stress more than lower it,” even when the work is useful; sunrunner attributes it to an interface that hands back more text than it takes in, which costs attention when nothing collapses neatly into an answer.
  • whiplash451 — the working compromise: choose the model by role. Coding agent, yes. Life adviser, no.

Where the thread pushes back on the premise

Not everyone accepts that dislike of the voice is a real finding. mahboi reads the personality differently — “they’re like 2020 FAANG leadership, which pisses me off even more” — and Retr0id files it under ordinary computing: “Computers frequently do things that annoy me, but I use computers nonetheless.” saadn92 reports the opposite of a cost: models replaced several paid apps for him. reedlaw reports being stuck with unpleasant prose because the model that produced the best results in A/B testing happened to be the verbose one, and settles for curating notes to blunt the worst tics. One commenter, cuzezzzbbfofai, suggests Fowler’s real complaint is obsolescence; javanissen answers with the essay itself — “the extremely clear reason he laid out in the article he wrote.”

On reading comments as evidence: HN handles are pseudonymous and the site publishes no per-comment scores, so the ordering is HN’s own ranking rather than a vote. This is a slice of a 224-comment thread, not a consensus.