Essay Argues AI Writing Tools Should Almost Never Be Used, Sparks 117-Comment Debate

An essay titled "Why You Should Almost Never Use AI to Write" by Erich Grunewald hit the front page of Hacker News this week, accumulating 212 points and 117 comments as of the post's run. The piece, published on Grunewald's Substack, takes a hard line against delegating prose to language models, and the comment thread quickly became a proxy fight over where the line between assistance and authorship actually sits.

Grunewald's argument, as framed by the title, is not that AI writing is useless but that its costs usually outweigh its benefits outside a narrow set of cases. The essay lands amid a steady stream of workplace surveys showing growing reliance on tools like ChatGPT and Claude for drafting emails, reports, and marketing copy, even as universities and publishers tighten disclosure rules. A Hacker News post reaching 212 points typically signals a topic that resonates well beyond the site's core readership of engineers and founders.

The comment count is the more revealing number. A 117-comment thread on a single Substack link means readers were not just nodding along; they were contesting the claim. Hacker News threads on AI writing tend to split into recognizable camps: developers who say the tools are fine for boilerplate and code comments, writers who argue the output flattens voice and introduces errors, and a third group that treats the debate as settled in favor of whatever ships fastest.

The timing matters. AI vendors have spent the past two years pushing writing assistance into every text box, from Google Docs to email clients, and the default is increasingly opt-out rather than opt-in. That shift makes an absolutist position like Grunewald's easier to dismiss as nostalgia and harder to ignore as a real objection. The essay's traction suggests a durable appetite for arguments that treat writing as a skill worth protecting rather than a task to be automated.

What the thread did not resolve is where the boundary should fall. Grunewald's title allows for exceptions, but the comments show no consensus on what they are: brainstorming, translation, editing, or nothing at all. For the AI industry, the practical question is whether users keep reaching for these tools by default or start asking what they give up when they do. The next data point will come from adoption numbers, not essays.