Sophie Alpert

There are no lossless transformations of natural-language text

June 25, 2026

In my work at Clay I recently wrote an internal policy on acceptable use of AI writing by engineers, and I’m sharing it here. It’s my hope that one day better AI tools might be able to help us think, but until then I fear that using AI to write does the exact opposite.

Good writing is a tool to clearly communicate ideas from your brain into someone else’s.

As of 2026, AI models — despite their coding ability — are not yet at a point where their unedited output will achieve this goal; you as an author need to take the time to make sure that all of the ideas in the writing are the ideas that you personally intend to convey (including the structure and wording that determines which ideas are emphasized) and that the documents are a good use of your readers’ time.

It’s allowed to use AI tools while brainstorming or drafting your writing and certainly while proofreading, but make sure to consider the following principles while doing so:

  • 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. In some cases, readers will recognize and call out the incongruous parts; in others, they will be misled as to what your actual thoughts are.

  • Writing is thinking. Spending time on the writing process — on deciding what to emphasize and how to structure your ideas clearly — teaches you more about your topic. If you circumvent this process, you will probably walk away with a poorer understanding of the subject matter. In many cases, written artifacts like tech specs, project status updates, and incident retrospectives serve as a “proof of thought”. The artifact itself is not the only goal; instead, detailed thinking about the problem is the goal. Outsourcing the creation of these documents to AI (so that you can skip the thinking) risks circumventing this very purpose. Even if working with AI helps you think through a problem, you’ll understand it better if you thoroughly review the result yourself.

  • More time should be spent authoring a document than consuming it. If you generate a document from a short prompt then ask your readers to go through the longer output, you are disrespecting their time. They can always talk to ChatGPT themselves if they want to. Most docs are written by one person but are read by many people, so any extra time that readers need to spend to understand what you meant incurs a multiplicative cost on the team’s time. Conversely, if you spend extra time to make your document clear and concise before sending it, you are paying a one-time cost that every reader will benefit from.

  • Longer is not better. Pascal once wrote, “I have made this [letter] longer than usual because I have not had time to make it shorter.” AI makes it much easier to generate a long doc, and one of its strategies is to include many sentences that don’t say much at all and detract from the actual content. If you are producing a longer piece of writing from a shorter prompt, consider instead just sharing the prompt itself. If you use AI to edit without lengthening, vacuous sentences are less of a risk but your meaning may still get obscured.

    There are no lossless transformations of natural-language text — every rewrite and rephrase changes the meaning of your writing, and if this is done by an entity that doesn’t have the most detailed mental representation of what you personally were trying to communicate, information will be lost. Readers will appreciate hearing your thoughts, even at the expense of supposed “polish”.

It’s also OK to quote AI generations verbatim that don’t meet the above standards if you mark them clearly as such. Sometimes it’s useful to say like “Claude offered this idea, do you think it’s worth looking more into?”, and this is allowed.

As AI tools improve over time at theory of mind and get better at writing, it may make sense to lean more heavily on them, but the principles above will remain important.