Within my industry, there has been a proliferation of AI-generated content being shared by humans for consumption by other humans. I have seen it everywhere, including technical design documents, press releases, bug reports, communications with customers, pull request descriptions, and even slack messages. I understand it too; good writing is difficult and time consuming, and it can seem like a distraction from the fun parts of the job like fixing and building stuff. There is a problem with delegating it though.

It is obvious. And the result is poor.

Writing is hard by necessity because reading is hard. In fact, reading is one of the most demanding activities we do. Because the human brain is not naturally suited to reading, it must rewire itself to facilitate it. It requires a long attention span and physical coordination. Research also shows that reading on a screen is even harder than reading physical material. As you can therefore imagine, reading something that is poorly written only exacerbates the challenge.

When you write for other people, it behooves you to do it well, so that your audience can comprehend your meaning quickly and accurately. This means organizing your thoughts cohesively, using a natural tone, pruning excess, and proof-reading. Conveying information is easy, but doing it concisely is an art and takes refinement.

LLMs don’t do any of these things, they simply spew. A yes-or-no question may incur paragraphs of qualifier text. They flood their responses with jargon and excessive punctuation. They overuse certain words and structures to such an extent that they sound out-of-place (e.g.: “this is a real signal”, “and that matters”, or the classic, “it’s not X – it’s Y”). Because all of the information is there, we sometimes make the mistake of accepting such text as-is, or with minimal changes. Our desire to get back to the fun stuff makes this practice more appealing.

But this hurts the reader and, by proxy, their relationship with you. They stumble through tedious ramblings, struggle to extract meaning from the fluff, and perhaps worst of all, they know that you didn’t take the time to make their experience better.

My stance on this is unequivocal: if you are sharing text content with another person, it should be yours. When using an LLM, you still must scrutinize the writing as much as you would any other text, and ensure the final draft is in your own words. Struggling to do so is an indicator that you shouldn’t be sharing it yet and may need to spend more time understanding the subject matter. If you lack confidence in your writing, LLMs should be used to help you improve your skills, not replace them.

Whether you find writing easy or hard, fun or boring, fast or slow, it is worth the effort. Other humans will thank you for it.