Sophie Alpert Warns AI Text Rewriting Is Never Lossless

Software engineer Sophie Alpert has introduced an internal policy outlining the acceptable use of artificial intelligence in technical writing, a framework recently spotlighted and endorsed by developer Simon Willison. The core of the policy addresses a growing reliance on large language models to polish, summarize, or restructure documentation. Alpert argues that engineers must remain entirely accountable for every sentence they publish, regardless of whether an AI assisted in its creation.

At the heart of this guideline is the assertion that "there are no lossless transformations of natural-language text," as Alpert writes. When an LLM is used to massage or rewrite a draft, the model inevitably alters the underlying meaning. Because the AI lacks a detailed mental representation of what the human author personally intended to communicate, critical information and nuance are lost during the translation process.

For practitioners, this policy shifts the responsibility of AI-assisted drafting back to the human author. Under Alpert's rules, engineers cannot excuse confusing or inaccurate documentation by blaming the AI. If a reviewer questions a specific line in a document, replying that an AI wrote it is deemed unacceptable. Writers must thoroughly review and stand behind every idea to ensure it genuinely represents their own thoughts before sharing it with colleagues or the public.

As generative AI tools become deeply integrated into software development lifecycles, this framework serves as a vital reminder of the limitations of machine translation in human communication. While LLMs can accelerate the drafting process, they cannot replicate human intent. Practitioners must treat AI writing assistants as fallible drafting tools rather than autonomous editors, ensuring that human oversight remains the final filter for all technical communications.