There is no such thing as an AI company.
At least not for most of us.
There are exceptions, and they're obvious. A handful of companies build the chips, the models, the infrastructure underneath all of this. Intelligence is their product. If that's your business, then yes — you are an AI company.
For everyone else, we're using the wrong language.
Every week I hear leaders describe their ambition to become "an AI company." Agencies. Software firms. Manufacturers. Professional services businesses. They all mean roughly the same thing: they want AI to become central to how they work.
The ambition is understandable. It's pointed in the wrong direction.
Almost none of us will build the underlying intelligence. We'll rent it. The models, APIs, copilots, and agents available to us will be available to everyone else — faster, cheaper, and more capable every year. Intelligence is rapidly becoming infrastructure.
And infrastructure rarely creates durable differentiation.
Electricity mattered. The companies built on top of electricity mattered more.
Cloud computing mattered. The companies that redesigned themselves around cloud computing mattered more.
AI will be no different.
You can't rent a moat.
Earlier this year I argued that most AI initiatives weren't failing — they were never alive. Organizations approached AI as a technology project when the real challenge was redesigning how the business worked. Since publishing that piece, I've had the same conversation over and over with founders and executives. The details change. The problem doesn't.
One founder I coach came back from a technology conference this spring and said:
"I'm not old. I'm running an old business."
His agency wasn't struggling. Revenue was healthy. Margins were strong. Clients were happy. The business had been founded with AI in its DNA. And it had quietly drifted back toward operating like a traditional agency.
Nobody made that decision. No leadership offsite concluded the company should become less innovative. It simply happened — one undocumented process, one exception, one ad hoc decision after another, until those decisions became the company's operating system.
That conversation didn't change my thinking. It hardened it.
Every organization already has an operating system. Some are explicit. Most evolved accidentally.
AI doesn't expose whether you've adopted new tools.
It exposes whether you've intentionally designed how you think and operate.
Which leads to the question that matters:
If every company can rent the same intelligence, what becomes scarce?
Not better prompts. Not more sophisticated agents. Not choosing the "right" model.
The scarce asset is organizational intelligence.
Not artificial intelligence — the accumulated knowledge of how your business makes decisions. Why one recommendation was chosen over another. Which framework consistently produces better outcomes. When experienced operators break their own rules, and why.
That's what clients actually pay for.
Professional service firms describe themselves by their outputs — research, campaigns, software, advice. But outputs are the visible artifact. Two firms can produce documents that look remarkably similar. What separates them is the accumulated judgment that produced them.
For decades, that judgment lived inside experienced employees, transferred slowly through observation and apprenticeship. When those people left, it left with them.
The best companies I've worked with all did one thing consistently: they turned individual judgment into organizational capability. What's changed is the economics: the cost of applying that judgment at scale has collapsed.
That's a much bigger opportunity than automating tasks.
Most companies start in exactly the wrong place. They begin with agents, because agents are visible. They can be demonstrated. Purchased. Announced at an all-hands.
Beneath every useful agent are three layers that actually matter.
Trusted information — where truth lives. The relationships between things — this client, that methodology, those conversations, these outcomes. And the judgment layer — the playbooks, quality bars, and decision rules your best people apply without thinking.
Only then do agents become powerful, because they stop improvising and start operating against knowledge the company actually owns.
Notice what compounds in that model.
Not the agent. Not the model. The organization's accumulated judgment.
Everyone is talking about capturing more data. Almost nobody is talking about capturing decisions.
Data is a record of what happened. A decision is a record of why you chose — with incomplete information, against alternatives, under pressure. The machine can process what happened all day long. It can't recover the why unless someone wrote it down.
Start treating decisions as assets. Why was this approach selected? Where did experience overrule the framework? What held up six months later?
The simplest version is a decision log. One file. Started this week. Every meaningful call: what was chosen, what it was chosen over, why — and what happened.
Those answers become the organization's memory — the judgment that used to leave when people left. New employees learn from it. Experienced employees build on it. AI systems reason across it.
That's where augmentation gets interesting. The machine carries the low cognitive load: it remembers every similar engagement, surfaces precedents, identifies patterns across years of work. People carry the high cognitive load: context, ambiguity, creativity, judgment — the decisions that define the business.
That's a partnership worth building.
It also forces a different conversation in the boardroom. Most organizations measure AI through activity: prompts written, hours saved, pilots launched, licenses deployed. Those metrics tell you whether people are using the technology. They tell you nothing about whether the business is changing.
The question is simpler: is the organization becoming structurally more productive?
One metric I keep coming back to is revenue per headcount. Not because it's perfect, but because it forces the question that matters most: what did we actually do with the capacity we created? Did we grow faster? Serve more customers? Enter new markets? Or did we just get more efficient at maintaining the business we already had?
And to be clear: cutting your way to a better ratio isn't transformation. The numerator is where the game is.
That's the distinction between adopting AI and redesigning a company around it.
Over the next decade, almost every business will become an AI-consuming business. Saying "we use AI" will sound as unremarkable as saying "we use the cloud."
The companies that pull away won't be AI companies.
They'll be better-designed organizations.
The vendors will keep advancing artificial intelligence. That's their business. Ours is different.
Because in the end, everyone will have access to the same intelligence.
Very few companies will own their own.