Wr-AI-ting

A series teaching you about AI, written for Product Managers (PMs). I’d recommend engineers give it a read (and subscribe!) too, as an easy way to stay abreast of the latest developments in the space. To point to a specific example, I enjoyed reading the history of LLMs chronicled here. Recording the present may seem mundane at times, but over the years it provides great value by allowing a kind of time travel to the past.

I know the author of this substack - she’s my cousin - and what I really want to talk about is writing. Specifically, about writing in today’s day and age where you have this personal assistant that will move your hand for you. She’d sent me one of her early posts and as I read it my interest slowly started to fade. I replied with the frankness only an elder brother can afford: The AI-fication of prose was prominent and staring me dead in the eye. My eyes, in response, glossed over, refusing to convey anything to my brain. I am not a fan of AI-written prose. To her credit, she took the feedback and the text stopped getting a Claude-branded regurgitation. Grammatical mistakes grew, and the articles were immediately more appealing.

The latest article in this series, for all its flaws and foibles, has some immense advantages in lacking AI-written slop that is slowly filling the world. For one, it conveys to the reader that this person stands behind what they have written. For all that they got right, I know there must be some level of underlying expertise or passion backing their text. For the things they got wrong, I know that they are at least brave enough to articulate what they think they understood and likely, open to being corrected and learning more. As Marc Brooker mentions in his Writing Is Magic blog, “Writing is nature’s way of letting you know how sloppy your thinking is”. Even if no one reads what you’ve written, you’ve grown merely by the act of thinking it through and writing it down. Second, it adds diversity. There are turns of phrase that I cringe at; that don’t fit my vibe so to speak. However, for every single choice of words I disagree with, there will be someone who is able to understand the author’s point better because their mind rides the same wave. You like your favorite authors not because they write perfect English, but because there is a certain core that runs through their writing that floats your boat.

I repeat, as Marc Brooker writes:

[…] asking people to read LLM-generated text breaks a kind of social contract. When I publish a piece of writing under my name (e.g. in my blog, or a document at work), I want the reader to know that I deeply understand and own what I wrote.

Specifically, it breaks the social contract. LLM-generated text is verbose, messy, poorly articulated and - to me - lacks the appeal of reading a human’s take on something. With AI-generated text, I struggle to connect the dots, to piece together what is actually trying to be conveyed. AI does not understand which lines are important nor which words and terms are confusing and need to be explained. Like Brooker says, if you’re going to send me AI-generated text, just send me the prompt. I have AI too and I can use it to shape it to my vibe!

Lastly, a huge gripe I have with AI-text is that it forces certain phrases and patterns. The human brain is an amazing machine, and it is able to spot these without explicit instruction (<insert 3-page article about “real” intelligence>) but on occasion, I find myself reaching into my bag of words and wondering if the color on them was because AI held the paintbrush. I’m sure there are times when - due to the amount of AI-text I’ve read - the color is all due to AI. The piece seems colorful to me, but I glance around and I see the same colors in pieces all around me. AI has us writing one way, and by seeking feedback from AI, we end up shaving away our uniqueness - mistakes and all - to showcase the exact same palette as the next person. I grew up with those around me singing along to Pink Floyd. Maybe the adoration was simply teenage angst causing us to spite our current masters and predicament, but somewhere along the way I think we internalized what “thought control” really meant. I look around me and within, and think of the days when we hoped not to be just “another brick in the wall”.

Appendix

Thoughts on AI-writing

I’ve decided to use this article as a collection point for any articles on AI-writing I find interesting, or from people whose writing or perspectives I value.

https://simonwillison.net/2026/Mar/1/ai-writing/ https://brooker.co.za/blog/2022/11/08/writing.html#foot1 https://brooker.co.za/blog/2026/06/18/my-blog-and-ai.html

AI-feedback

I’ve run this post through some LLMs and they’ve pointed out “flaws”. For each instance, I first tried to guess what the flaw would be by reading the text referenced. If I spotted a flaw, I corrected it. If I didn’t, I took a look at the AI’s suggestion and considered it. I did not seek feedback on phrasing my text differently. Feedback was only on punctuation, typos and grammar (using Simon Willison’s prompt linked above). At the end of a few iterations, AI still had many mistakes it wanted me to fix. Things it found jarring or logically inconsistent (despite my prompt!). I did not change these. In the times before LLMs, my mother used to proofread all my posts. Now, with LLMs, I should be glad to absolve her of the chore, but I don’t think she saw it as such. It was simply an opportunity to connect and help her son. This convenience the LLM creates takes away an opportunity for a social connection. Wiith friends, there is no wonder anymore, there is only “let me ask the LLM”. At work, there is no need to have the architecture anymore, there is only “let me ask the LLM”. For every second of grunt work saved, it has taken away minutes of social interaction that brought me joy. However, the choice is still with me. And even with the “smartest” LLMs around, I’ll choose to send this post to Amma, seek her feedback and dear reader, you can know that it is her feedback that has shaped this.