Here’s something that’s been nagging at me lately.
AI has become an amazing writer. Truly amazing. Give it a prompt, and it comes back with coherent prose, interesting turns of phrase, even a sense of rhythm and flow that would have made my high school English teacher proud.
And here’s the thing—it captures my voice remarkably well. Feed it enough of my writing, and suddenly it’s crafting sentences that sound like they came straight out of my head. The em-dashes are in the right places. The rhetorical questions land where they should. It even knows when to throw in a short paragraph for emphasis.
Like this.
So you’d think this would be a godsend, right? That I’d be cranking out content at lightning speed, freed from the tyranny of the blank page?
Not exactly.
The Problem With Perfect
The problem isn’t that AI writes badly. The problem is that it writes too well—in too many different ways.
When I ask ChatGPT to draft something, I get one version. Ask Claude, and I get another. Hell, ask the same platform twice, and I’ll get a third version that’s equally compelling but completely different.
And each one has moments of brilliance.
Version A has this perfect opening hook that grabs you immediately. Version B nails the middle section with an analogy I never would have thought of. Version C sticks the landing with a closing line that’s pure gold.
So naturally, I think: “Why choose? I’ll just combine the best parts from each version and create the ultimate ‘greatest hits’ edition.”
Sounds logical, right?
Enter Frankenstein’s Monster
But here’s what actually happens.
I start cutting and pasting. Moving paragraphs. Trying to splice together the best opening from Version A with the killer middle from Version B and that perfect closer from Version C.
And suddenly—despite having assembled all the best parts—I’ve created something that feels... off.
The narrative flow is gone. The transitions are clunky. What worked beautifully in context now feels forced. I’ve got all the right ingredients, but somehow the dish doesn’t taste right anymore.
So I start editing. Smoothing transitions. Rewriting connections. Adjusting the tone so it feels consistent throughout. Tweaking word choices. Removing repetitions that emerged when I merged similar sections.
And then—this is the truly maddening part—I look up and realize I’ve spent more time wrestling these AI versions into coherence than it would have taken me to just write the damn thing myself from scratch.
The Cutting Room Floor Blues
But wait—there’s more!
Even after all that work, I’m not left with a sense of satisfaction. I’m left with a vague feeling of disappointment.
Because buried in those other versions—the ones I didn’t use—were brilliant lines, clever transitions, and interesting angles that I had to abandon. Great stuff that deserved to see the light of day.
It’s like being a film editor who has to cut three amazing scenes because they don’t fit the final narrative, even though each one is objectively excellent on its own.
When I write something longhand, I make choices as I go. Each sentence informs the next. The piece develops organically, and I never see the roads not taken. What you read is what emerged from the process.
But with AI, I’m constantly haunted by the ghost of alternate versions—parallel universes where I made different choices and ended up somewhere equally good but completely different.
The Lesson I Should Have Learned Long Ago
You know what the real answer probably is?
Don’t let perfect be the enemy of the good. Just pick one version and hit publish. Move on.
It’s like running a startup, actually—you can spend forever refining your product, looking at what competitors are doing, trying to incorporate every great feature you see. Or you can ship what you’ve got and iterate based on real feedback.
The poet Paul Valéry wrote something in 1933 that keeps coming back to me: “In the eyes of those who anxiously seek perfection, a work is never truly completed—but abandoned.”
He was talking about poetry, but he might as well have been talking about AI-generated content.
There is no perfect version. There’s just the moment when you decide you’re done—when you choose to abandon the pursuit of something better and accept what’s in front of you as good enough.
Maybe that’s the real skill we need to develop in this AI age. Not the ability to perfectly merge multiple versions into one ideal piece. But the discipline to accept one good version and move forward.
Even if it means leaving other great options unexplored.
I use it only for polishing and to keep my tone and style.
With clients, my work is about “cutting the noise”. In AI, people obsess over features and fireworks instead of what actually matters.
Today, finding less is more valuable than producing more features
I started using AI less and less for anything that's writing, and more for what's behind the scene.
Brainstorming, polishing, criticizing stuff, and so on. It works best as an objective assistant.
But no matter how hard you try to make AI write just like you, it can't nail your voice in a human way, it'll always be too perfect.
And I'd rather have something flawed that reflects my mistakes, or my flawed thinking than something that's not me.
Interesting piece though, but I won't ever agree with copy-pasting posts from the LLM straight to my LinkedIn... people could read that in their GPTs...