A few months ago a client forwarded me a brand manifesto and asked what I thought. It was clean. Grammatical. It hit every beat you would expect — bold, human, relentless, the usual trinity. It was also completely dead. Reading it felt like shaking hands with a mannequin. Warm enough at a glance, but nobody home.
The machine wrote it, of course. And the machine did nothing wrong. It did precisely what it was asked: produce competent words about a company that wants to sound brave. The problem is that competent and brave are different jobs, and only one of them can be automated.
I have spent my career being paid to find the right words when the direction is blurry and the stakes are high. So you might expect me to be nervous right now. I am not. I think the people writing brand copy by hand are about to become more valuable, and I think the reason is hiding in plain sight.
The cost of words just went to zero. The cost of judgment didn't.
For as long as anyone reading this has been alive, writing was bottlenecked by writing. You wanted ten headlines, you paid for the hours it took a person to make ten headlines. That bottleneck is gone. Today you can have a hundred headlines before your coffee is cool.
But a hundred headlines is not a gift. It is a new problem. Somebody still has to know which one is true to the brand, which one a competitor already owns, which one will sound clever today and embarrassing in a year. The machine has made production free and made taste the entire game.
The machine made production free. Taste is now the entire game.
This is the part people miss when they panic about automation in creative work. The threat was never that the machine would have ideas. The threat is that it would flood the room with plausible ones, and that we would lose the nerve to throw the plausible ones away.
What a brand actually buys
When a company hires someone like me, they think they are buying words. They are not. They are buying a series of small, high-conviction decisions about who they are. A voice is just the accumulated residue of a thousand of those decisions — what we say, what we refuse to say, the joke we make instead of the one everyone expects, the adjective we cut because it was trying too hard.
A model trained on the entire internet is, almost by definition, the average of everyone. That makes it a phenomenal intern and a terrible point of view. It will give you the most likely next word every single time, and "most likely" is another way of saying "what everyone else would have written." For a brand trying to stand out, that is the one thing you cannot afford.
So what do you actually do with it?
You use it the way a director uses a camera operator. The machine is astonishing at the parts of the work that are labor: variations, first passes, summaries, the forty alternate subject lines, the research that used to eat a Tuesday. I would be foolish to do that by hand now, and I don't.
What I do not hand over is the part where you decide what the thing is for. The angle. The single risky sentence the whole piece is built to deliver. The choice to be quiet when the brief begged for loud. That work has never been about typing speed, and the tool that types fast does not touch it.
The writers who struggle in the next few years will be the ones whose only skill was assembling correct sentences, because correct sentences are now a commodity. The writers who thrive will be the ones who were never really selling sentences at all. They were selling judgment, and judgment just became the rarest thing in the room.
Why I'm writing this here
This is the first piece in what I intend to keep as an ongoing notebook — about brand storytelling, the craft of voice, the strange and fast-moving present of this industry, and occasionally just what I'm learning by doing the work. No schedule of hot takes. No SEO sludge. Just the thinking I would do anyway, written down where someone might find it useful.
If you make brands, lead creative teams, or just care about why some words land and most don't, I hope you'll stick around. The machine can write the average of everything. I'm more interested in the things it can't.
