Most writing is about to get worse. The models are fluent, and fluent is easy to mistake for finished. But the words were never the work. The thought is the reason you are writing. The words are only the transport. When the transport gets sloppy, people stop trusting the thinking behind it, and they are usually right to.
This is the thing the executives I work with keep saying out loud. They are tired of reading AI slop dressed up as analysis. They do not want a model outsourcing their judgment, and they do not want their own thinking flattened into the same beige paragraph everyone else is publishing this quarter. The answer is not to put the tools down. It is to refuse to let them think for you.
The tell with AI writing is not grammar. It is sameness. It hedges. It pads. It reaches for the same two hundred words, the ones every model loves and no person says out loud. Read enough of it and you stop reading. The machine is fluent and forgettable.
I spent the last year on the opposite problem: getting a model to write in one exact human voice. Mine.
So I built an engram, a profile of how I write. It knows my sentence lengths, the words I reach for, and the ones I refuse. It knows where a parenthetical earns its place, and how I open an email, which is never with "I hope this finds you well." I pulled it from around five hundred of my own messages, emails, internal docs, and articles. Then I handed it to the model as the standard to hit, not a suggestion to consider.
The resume on this site was generated the same way everything here is. I kept cutting until it stopped sounding optimized. If it reads like a person, that was the hard part. You can decide whether it worked.
Read the engram ↗ See the resume →
The model is not the writer. I am. This is editing, not engineering. The model gives me options. Judgment keeps the one line that sounds like me and kills the three that sound like a press release. That part is not in the tool. It comes from knowing how you sound, and refusing everything that doesn't.
Now I'm running the same craft at length. I'm writing a book for middle managers. Keeping one paragraph in your voice is easy. Keeping a whole manuscript in it is not. So I built book-writing skills to hold the line across that distance. They check whether a chapter argues something or just piles up related points. They flag where the voice slips and tell me what to cut. The model doesn't write the book. I do. It just keeps the work moving.