Fowler Warns Writers: Don't Let LLMs Draft for You
Original: LLM Voice
Why This Matters
Signals growing pushback from influential technical voices on AI-assisted writing norms.
Martin Fowler argues that LLM-generated prose creates an "uncanny valley" effect in readers, urges authors to write all drafts themselves, and warns that delegating writing to AI also delegates thinking.
In a post on martinfowler.com, Thoughtworks chief scientist Martin Fowler describes a growing unease when reading text he suspects was generated by an LLM — what he calls "LLM-Voice." He argues the reaction is inherently human: a sense of uncanny valley in prose that no detection tool can fix after the fact, because the discomfort arrives before any tool is reached for.
Fowler is careful not to conflate LLM-Voice with generational language shifts. Phrases like "adopting a new sales motion" irritate him, but he acknowledges hearing them from humans, not just AI outputs.
His practical advice is pointed: never use an LLM to write a first draft. Even heavy human editing, he says, rarely purges the LLM-Voice fully. Accepting rephrase suggestions is also risky. Using AI for high-level outlining may be acceptable, but authors must write every sentence themselves. He notes one exception — a piece he knew was LLM-drafted showed no detectable AI voice in the final version, but only after a professional editor at a major publication overhauled it.
Fowler's deeper concern goes beyond prose style. Writing, he argues, is a tool for thinking. Outsourcing drafts means outsourcing the cognitive work of forming ideas. His own AI use is limited to search — and he verifies every result at the source.