A clear document gives another person a way to follow the writer’s thinking. Martin Fowler asks what happens when producing the sentences becomes easier than doing that thinking.

In his October 29 essay, Fowler describes his aversion to prose that sounds generated by a language model. He is wary of automated detection and doubts his own ability to identify the source reliably. His concern is the reader’s response and the writer’s understanding, rather than proving who produced a passage.

Fowler recommends that authors write their own prose, arguing that editing a generated draft often leaves its characteristic voice behind. He also describes writing as a means of resolving confusion: working through the words helps him discover what he actually understands.

This is a practitioner’s argument, not an experiment demonstrating that every use of writing assistance reduces comprehension. His position includes an accessibility exception for Jim Highsmith, whose Parkinson’s disease makes AI assistance valuable to his writing. Fowler treats translation as a more difficult tradeoff and acknowledges situations where communication with a process may justify generated prose.

For a team preparing architecture decisions, incident reports or project proposals, the practical question is whether the accountable person can explain the document. A review can ask for the decision, its supporting evidence, the rejected alternative and the remaining uncertainty in ordinary conversation. Difficulty answering those questions identifies work the polished draft has not completed.

That review should preserve assistance that lets someone participate. The tool’s role and the person’s ownership can both remain visible without treating one writing method as suitable for everyone.

Before approving an important document, ask its owner to walk through the choice it records. The explanation should survive after the finished sentences are put away.