AI made 40 versions of your company and you didn’t even notice

Everyone got faster, and somehow the company got blurrier.

Two good marketers on my team sent me social posts for our company in the same week. Both nailed the assignment. For some other company. Not ours. Nobody got it wrong, which was the problem.

One used a prompt they wrote last month. The other copied one from some self proclaimed AI guru who sounded very confident on YouTube. Both answers came back plausible and professional. Both were useful enough to move the work forward. Neither gave us the same company back.

That is the part of AI adoption that feels small until it compounds.

Give people better tools and they move faster. The blank page gets less blank. The first draft arrives in minutes instead of hours. The person who used to wait on someone else can now get to a decent version alone. That works. That is why AI is so easy to sell.

But inside a company, speed is only half the job. The harder question is whether everyone is still working from the same idea of what good looks like here. Not in theory. Here. And AI, used casually, is very good at pulling that apart.

I have watched it happen inside the same team. Same brand, same goals, two people, different prompts. Different tools. Different private workspaces. Different versions of what the company sounds like. No one is being dumb or reckless. Usually the opposite. People are just trying to get on with the work.

So the company paid for intelligence and accidentally cloned itself forty times. And the companies that didn’t pay? Their people are using AI on the sly anyway, swapping prompts like contraband and smuggling the good stuff in through the side door. Either way you land in the same place with forty sole traders sharing your logo.

I think this is where a lot of teams misread AI adoption. They treat it like personal productivity software, and at first that is exactly what it looks like. You get the email written before lunch, the messy spreadsheet explained back to you in plain English, the deck you were dreading knocked into shape in twenty minutes. The small stuff that used to sit in the swamp between meaning to do something and actually doing it. So everyone starts where the reward is obvious. Their own work.

The bit people underestimate is that a company is not a stack of individual outputs. It is a shared operating system. Usually a messy one, held together by old decks, half-remembered decisions, the elders who know why the product is called that, and the one in marketing who can tell which sentence legal will hate before legal has seen it.

And here is where it starts to break. When the clever prompt lives in one person’s head and one person’s chat history, the capability belongs to them, not the company. It walks in with them in the morning and walks out with them when they leave. What is left behind is a folder nobody else can quite reconstruct.

The part that excites me is that AI could do the opposite.

Used well, the same tool that splinters one company into multiple personalities can become its central memory. The brand, the standards, the product truth, the customer history, the awkward internal scars. Available to everyone instead of locked in whatever prompt someone happened to write that week. The thing that worked lives somewhere shared, not in one person’s chat history.

The new person does not start from zero. They inherit the company. Which sounds noble until you realise most companies store their actual operating knowledge in seven Slack threads, three people’s heads, and a Google Doc last updated during Covid.

Not the polished version from the brand deck, but the real version. The good, the bad, and the weird internal logic nobody ever writes down.

They inherit the sentence legal always kills. The claim the founder hates. The customer story sales keeps using because it works. The product detail everyone gets wrong until someone from product corrects it.

The stuff that never makes it into the onboarding deck because everyone assumes everyone else already knows it. Which is how companies end up treating folklore like infrastructure.

That is the real asset, the company memory.

Because if AI cannot see how the company thinks, it will guess. And it will guess with confidence. But that is a different job from buying tools and hoping clever people use them well. The good prompts need to live somewhere the whole team can reach. The brand rules need examples. The product truth needs owners. The customer stories need to be reusable. Otherwise the output stays a lottery.

The pattern feels familiar from early digital transformation. Everyone assumed the technology was the hard part. Then the technology arrived, and the hard part turned out to be people, habits, coordination, and moving the oil tanker without pretending it was a speedboat. AI has the same shape. Only faster. More abstract. More hyped.

Everyone wants the silver bullet. And the quick wins are real, which is what makes this awkward. The founder who says give people the tools and let them run is not completely wrong. I know, because I said it.

For a while I believed it. Give good people good tools and get out of the way. Then I watched good people produce more and more work. Faster work. Work that looked right until you held it up against the company. None of it was ours.

Over time, the question changes. Did the company get smarter, or did forty people each build a patched up machine that walks out the door when they do? That is the difference between rolling out a tool and changing how the place actually works.

One makes people faster immediately. The other makes the company smarter slowly.

Most teams will confuse the two.

See you out there

Martin

P.S. This week’s soundtrack “Everything In Its Right Place” by Radiohead.