Martin Fowler loses interest when prose starts to sound generated by a large language model. He doubts his ability to identify it reliably, but a detector’s verdict cannot restore his desire to read. His larger worry is what happens when authors delegate the work of finding their own words.
- Editing an AI-written first draft can leave its voice intact. Fowler recommends writing the prose yourself, even if a model helps with an outline.
- Writing helps him discover what he thinks. Handing that struggle to a model may also hand over the work of understanding.
- Accessibility changes the trade-off: AI helps his friend Jim Highsmith keep writing despite Parkinson’s disease. Translation and routine form letters are further cases where assistance can make sense.
This is a writer’s argument about practice and communication, not a study measuring lost thinking skills. Fowler’s own policy is to use AI for search and verify what it finds at the source.
The 152-comment thread on Hacker News adds a practical alternative and disagreement about where to draw the line.
What the thread adds
- tptacek — use the model as a critic while keeping authorship of the sentences: “Better, by a lot, to prompt the LLM not to suggest any words, but just offer objective critiques.”
- hectdev — broadens the accessibility objection to learning disabilities and language fluency: “High literacy such that you can discern LLM Voice is quite an accomplishment but blanket dismissal reeks of privilege.”
- cadamsdotcom — accepts generated wording with substantial revision, but draws a personal boundary around first-person authorship: “If you use the first person pronoun “I”, you better make damn sure you wrote those words.”
Handles are pseudonymous; HN publishes no per-comment scores. The sampled order is HN’s ranking, and this is a slice of the thread, not a consensus.