A Culture of AI Is Just How Your People Talk About It

Every company I talk to says the same four words: we need to do AI. Almost none of them can tell me whether the culture underneath that sentence is alive or dead — and that, far more than the tooling, is what decides whether any of it works.

I’ve been circling this idea for a while. It started as a phrase I’ve always liked — a culture of experimentation — and it kept resolving into something more urgent. Not the budget. Not the security review. The smaller, more human layer underneath all of it: the way your people actually talk about AI.

What a culture of AI actually is

When leaders hear “do AI,” they reach for the big levers — licenses, access, a policy PDF, a cybersecurity review. Those matter. They’re also not where teams are losing.

You’re losing in the tone. If using AI is high-status on your team — something people do in the open, compare notes on, get visibly better at — the skill compounds fast. If it’s low-status — something you’d be a little embarrassed to admit in a meeting — people use it in secret, use it badly, or don’t use it at all. Same tools. Opposite outcomes. The variable is culture.

A culture of AI is mostly permission — to use it, to be a beginner at it, to say “the model wrote the first draft” without losing standing in the room.

The culture of experimentation, and the one we need now

Years ago I read about companies that built a culture of experimentation on purpose. Not a slogan — systems. They engineered the safety to run a test, fail it, and say so out loud, because learning only compounds when nobody’s ashamed of the miss. Fear was the thing they had to design out.

AI is that same problem with the volume turned up. The tools change every week. Nobody is really an expert. Everyone is a beginner again at the same time — which is rare and valuable, and easy to waste. A culture that punishes visible beginners — with contempt, with the em-dash sneer, with “that’s just slop” — guarantees that no one gets good in the open. You end up with quiet, private, wildly uneven skill instead of a team that levels up together. (And the contempt is a little absurd when more than half of long-form professional writing is already machine-assisted — people are sneering at something most of them quietly already do.)

How to tell if your culture of AI is working

You can’t see a culture directly. But it leaves fingerprints. Here’s how to read yours — no survey required.

Healthy signs it’s working

• People share prompts unprompted — “here’s the one that finally worked.”
• AI comes up in meetings as a normal tool, not a confession.
• Someone says “the model drafted this, I rewrote it” and nothing about their standing changes.
• Beginners ask “how did you get it to do that?” without embarrassment.
• Failures get shared: “I tried it, it hallucinated, here’s what I learned.”

Warning signs it’s stalling

• The dominant tone is contempt — slop, cheating, count the em dashes.
• People hide that they used it, or launder the output so no one can tell.
• Only a few “AI people” ever touch it; everyone else waits to be told.
• There’s a policy but no practice — a document, not a habit.
• The loudest voices in the room are the most dismissive.

How to improve it

Culture feels immovable until you notice it’s built from a handful of small, repeatable signals. Change the signals and the culture follows. Six that actually move it:

1. Model it from the top, visibly. When the highest-status person in the room uses AI in the open — “here’s the prompt, here’s where it was wrong” — everyone else gets permission in a single move. Culture is set by what leaders do, not what the policy says.
2. Make it safe to be a beginner. Everyone is one right now — say so out loud. The person fumbling with a new tool in public is doing the most useful thing on the team, and they should feel like it.
3. Build a place to share prompts, wins, and misses. A channel, a weekly show-and-tell, a shared doc. The compounding happens when one person’s discovery becomes everyone’s in a day instead of a quarter.
4. Retire the contempt. Name the em-dash sneer and the reflexive “slop” for what they are — culture-killers. You can hold a high bar for the output without shaming the tool. Standards are about quality; contempt is about status.
5. Keep the human in the loop — loudly. The goal was never to hide the AI. It’s to stay in the loop hard enough that the work still carries your judgment, your stories, the thing only you would say. Make “augment, don’t outsource” the norm, and the pride point.
6. Fund the access and protect the time. Permission is free; the tools and the hours to learn them are not. A culture of AI that doesn’t pay for the seats or defend the learning time is just a poster on the wall.

My own version of this is small and daily. I use these tools every day, my writing carries the tells, and I say so — because I’m the human in the loop, and the loop is the entire point. What I don’t do is let the work flatten into something anyone could have typed. That’s the standard, and it has nothing to do with hiding.

A culture of experimentation taught people it was safe to be wrong out loud. A culture of AI asks for the same courage in a new key — safe to be a beginner out loud. The contempt is the only thing really standing in the way. And contempt is a choice you can stop making tomorrow morning.

— Joe