Against the slop machine.

There is a particular kind of AI draft that makes my teeth itch.

You know it immediately. The prose is smooth in the way hotel lobby music is smooth. Every paragraph is balanced. Every claim has been wrapped in soft corporate gauze. The thing is always being transformed, elevated, unlocked, or positioned at the intersection of 2 nouns that should never have met.

Nobody is angry. Nobody is funny. Nobody has a memory. Nobody has ever eaten lunch too fast over a keyboard because the deck was due at 3.

The text has no fingerprints.

This is the slop machine.

I say that as someone who builds AI systems and uses them every day. I am not standing outside the restaurant yelling about seed oils. I am in the kitchen. I know what the tools can do. I know how useful they are when the process around them is serious.

That is why the bad output bothers me so much.

AI is extraordinary at producing language that appears finished before it has earned the right to be read. It can summarize, reframe, draft, classify, compare, outline, and remix at a speed that still feels fake even after you have watched it happen 500 times.

The danger is that finished-looking text creates managerial hypnosis.

There it is. A paragraph. A headline. A campaign brief. A sales email. An executive summary. It has punctuation and everything. Someone can paste it into a doc, move the ticket, and feel the little narcotic glow of progress.

Then the text goes out into the world and sounds like it was written by a committee of microwaves.

The problem is not that AI cannot write. It can. The problem is that AI is usually asked to write without a point of view, without source discipline, without taste, without memory, and without fear of embarrassment.

Fear of embarrassment is important. It keeps people from saying "seamless end-to-end solution" in public.

A human writer has shame. A human writer knows when a sentence sounds like a LinkedIn post trying to sell a webinar to a regional sales manager. A human writer has heard a real person say, "Nobody talks like this," and felt the sting.

A model has no sting. It has probability.

That does not make it useless. It means the human has to bring the scar tissue.

When I built AI workflows for GTM work, the part that mattered most was not the prompt. The prompt mattered, of course. Prompts are the visible magic trick, so everyone stares at them. The real work was the governor around the prompt: source packets, claim rules, voice standards, approval gates, examples of good writing, examples of bad writing, and a willingness to throw away fluent garbage.

Especially the fluent garbage.

Fluent garbage is dangerous because it does not look broken. A factual error announces itself eventually. A broken link gets caught. A hallucinated source can be checked. But dead prose can live forever. It can colonize decks, emails, webpages, RFP responses, sales scripts. It can slowly teach a company to sound like nobody in particular.

Then one day the brand voice is gone. In its place: vapor.

I care about this because language is not decoration. Language is how a company reveals what it understands. If the writing is vague, the thinking is usually vague. If every claim sounds inflated, the proof is probably weak. If every sentence has that weightless AI sheen, I assume the work underneath it has not been inspected either.

This is a little unfair.

It is also usually right.

The better use of AI is quieter and much more useful. Use it to pull the source material together. Use it to find contradictions. Use it to test structure. Use it to make a bad first draft that gives the human something to push against. Use it to check whether the argument holds. Use it to catch the 19th instance of a phrase you have started repeating because your brain is tired.

Then write like a person who has been somewhere.

That is the standard I want in my own work. Not perfect prose. Perfect prose is often sterile. I want writing with judgment in it. A little pressure. A little blood. A sentence that sounds like it came from a person who had to make the call and live with it.

AI can help get there. It can carry some of the load. It can save hours. It can make small teams feel like they have a little more oxygen.

But it cannot care if the sentence has a pulse.

That part is still on us.