I'm not AI, I'm autistic

I replied to an email a while back and got a sarcastic one straight back. Thanks for using AI to answer me. I spent ages writing mine.

My reply was five words. It's not AI. I'm autistic.

Here's another one, which happened to somebody I know. They wrote a leaving message to a colleague they properly admired. Sat with it. Thought about what that colleague had actually done for them over the years, and how to say it without being naff. Pressed send.

Oh, nice AI.

I've not been able to leave either of them alone since.

Have a think about what a strange thing we've invented here.

You can now question somebody's sincerity at zero cost to yourself, and they have no way of disproving it. There's no receipt. No version history that counts for anything. Nothing anyone can produce that settles it.

The only defence available is to explain yourself. Which means telling somebody you barely know how long you spent, or why you write the way you do, or how your brain works.

(Sure there’s Claude watermarks and obvious Claude-ism’s but it’s not fool-proof, and we’re dealing with lots of fools)

That's what happened to me. To defend an email I'd written myself, I ended up telling somebody I barely knew that I'm autistic. He got to be clever for about eight seconds. I got to disclose a diagnosis.

I'm not after sympathy for that, it's a small thing and I got over it*.

But it's worth noticing what the accusation actually costs, and who ends up paying it.

*I’m of the belief that autism is a super-power and us autismos are actually better than you muggles 😉

Careful writing and generated writing throw off a lot of the same signals. Complete sentences. Punctuation that's correct. A structure somebody thought about. Nothing dashed off in ninety seconds between meetings.

So the people getting accused are:

That last group is the one that gets me. The leaving message got flagged because effort is now suspicious. Trying hard throws off the same signature as not trying at all.

If you’ve seen any of my talks this year at the various conferences I’ve been at, you’ll know that the “perception” of AI errods trust. Perception of AI, not AI itself, just the perception.

Here's the part that should piss you off.

Everyone is now prompting the thing to sound casual. Lowercase openings. Sentence fragments. The odd deliberate typo. Make it sound human, mate.

Which means the tells we're all pattern matching against are last year's tells. Our brains have wired the em dash to slop and there's no undoing that now. So we're catching people who own a copy of Fowler's and waving through people who typed one extra line into a prompt box.

I know people who now go through their own writing swapping em dashes out for the smaller one. Editing correct punctuation out of their own sentences to avoid being accused of something they didn't do. That's genuinely where we've got to.

There are no em dashes in this piece and I'd like that noted.

If you run a team, four things.

Stop saying it. Start from the assumption that they used AI, because it's 2026 and of course they did. Everybody does. Accusing somebody of using AI is like accusing a builder of using a hammer. Nobody cares about the hammer. If the house is wonky, tell them the house is wonky.

So when something reads like slop, your actual objection is that it's bad. Say that. "This doesn't say anything" is feedback a person can do something with. "This is AI" is an accusation they can only answer by defending their character, and when you're wrong you've done real damage to somebody who was trying.

Watch who keeps getting it. If the same person on your team gets pulled up for this repeatedly, go and check whether the work is genuinely weak or whether they just write formally. I'd bet on the second more often than most managers would like to admit.

Don't write a policy that bans the good use. Somebody I know doesn't think he expresses himself well in writing. So when he had to contribute a section to an article, he fed the model a pile of his own podcast episodes, got it to learn how he actually talks, then used that to draft it and edited it himself. What came out sounded like him. Arguably more like him than what he'd have managed cold. A blanket no-AI-writing rule bans that, and it bans it hardest for the person who needed it most.

Test whether they can defend it. The useful question was never whether AI touched the thing. It's whether the person who sent it understands it well enough to answer the next question about it. That test catches genuine laziness and leaves everybody else alone, which is more than any of the tells manage. (Go and Google “Meat Proxy” for a laugh)

The optimistic version is that we all eventually relax about it and drop the policing. I don't buy it, and nobody I've put it to buys it either. The push seems to go the other way, towards a premium on visible humanity.

Except there's a version of that future where visibly human just means deliberately scruffy, and we've all trained ourselves to write worse in order to prove we're real. Lowercase, no punctuation, half a thought. Certified organic.

I don't know what to do about that one.

What I do know is that the cost of getting these accusations wrong isn't being paid by the people churning out slop. It's being paid by people who did nothing except write properly.

So if you're about to fire off a "nice AI" at somebody, have a think about whether they spent their evening on it. Or whether you're about to make them tell you something about themselves they'd rather have kept to themselves.

Say what's wrong with the work instead. It's better feedback anyway.