Glonce

Simon Willison shares Sophie Alpert's AI-writing policy

miscThe Glonce editorsSimon Willison🔥59Glonce rating59Impact45Novelty50Relevance92Surprise40Players60Credibility72

Simon Willison wrote a short post sharing and praising Sophie Alpert's "internal policy on acceptable use of AI writing by engineers," calling it a good, short read that follows its own advice. He quotes two passages from Alpert's policy. The first sets the accountability rule he singles out as crucial for anyone who lets an LLM help massage their writing: an engineer must stand behind every idea and every sentence in their docs, and is responsible for making sure the whole document represents their own thoughts before sharing it. If a reviewer asks what a line meant, answering "Oh sorry, AI wrote that, just ignore it" is not acceptable, because it confuses readers and wastes their time with material that does not genuinely represent the author's thinking. The second passage expands on the idea behind the post's title: there are no lossless transformations of natural-language text. Every rewrite or rephrase changes the meaning of the original writing, and when that rewriting is done by an entity that lacks the most detailed mental picture of what the author personally meant to communicate, information gets lost in the process.

Key facts

  • Simon Willison shared Sophie Alpert's internal policy on acceptable use of AI writing by engineers, describing it as short and "really good."
  • The policy's core rule: an engineer must stand behind every idea and sentence in a document and is responsible for it representing their own thoughts before it is shared.
  • Under the policy, telling a reviewer "AI wrote that, just ignore it" is not an acceptable response to a question about a document's content.
  • The policy argues there are no lossless transformations of natural-language text: any rewrite by an entity without the author's full intended meaning loses information.
  • Willison frames the accountability rule as the crucial one for engineers who choose to let LLMs help with their writing.

Why it matters

Engineers increasingly use LLMs to draft, edit or "massage" technical writing, and the accountability for that writing tends to get fuzzy once a model has touched it. Alpert's policy, as Willison presents it, draws a hard line: the human author remains fully responsible for every sentence regardless of how it was produced, and the underlying reason is structural rather than a matter of discipline. Any rewrite by a party that does not share the author's exact intended meaning necessarily loses or shifts some of it, so an AI-smoothed document cannot be treated as interchangeable with what the author would have written unaided.

Who it affects

Engineers and other writers at organizations that let staff use LLMs to help with documentation, reports or other written work, along with the reviewers and readers who rely on that writing being an accurate account of the author's own thinking.

How to use it

The concrete takeaway is a policy engineers can hold themselves to: read and own every line an LLM helps produce before sharing it, and be ready to explain and defend any sentence in a document as if you had written it entirely yourself. Deflecting a reviewer's question by blaming the AI is explicitly ruled out.

How solid is it

This is a short opinion post on Simon Willison's personal blog, endorsing and quoting from a policy document written by Sophie Alpert. Willison quotes two passages directly but does not provide the full text of the policy or independent evidence of how it works in practice; the post rests on Willison's endorsement of Alpert's reasoning rather than on external data.

Risks and caveats

The source does not say where Sophie Alpert works or what her professional title is, does not give a URL or date for the policy document itself, does not say whether the policy is public or was shared privately with Willison, and gives no detail on how "acceptable use" is defined beyond the two quoted passages.

“You must stand behind every idea and every sentence in your docs. It is your responsibility to make sure that the entire document is representative of your own thoughts before you share it. If a reviewer asks, “What did you mean by this line?”, it’s not acceptable to reply with “Oh sorry, AI wrote that, just ignore it.” You will confuse your readers (and waste their time) if you present them things that are not genuinely representative of your thoughts.”

— Sophie Alpert's internal AI-writing policy, quoted by Simon Willison