Most companies are adopting AI the way they once adopted a website — because everyone else is. The ones that will win are asking a harder question: where does this actually change our economics?
There is a version of artificial intelligence that lives in press releases and quarterly calls, and there is a version that quietly changes how a business makes money. They are not the same, and confusing them is expensive.
The headline version treats AI as a signal: something to announce, a box to tick, a way to look modern. The operating version treats AI as a lever: a specific place where a task becomes faster, cheaper or better, and where that improvement compounds. When I work with companies on AI, my first job is usually to move them from the first framing to the second.
Start where the economics live
The wrong question is "how do we use AI?" The right question is "where in this business does a small improvement create a large effect?" Those points are where AI is worth pursuing. Sometimes it is a support function drowning in repetitive requests. Sometimes it is an underwriting or pricing decision made thousands of times a day. Sometimes it is the simple ability to read every document a company receives instead of a sample.
AI applied at these leverage points does not just cut cost — it changes what the company can profitably do. That is an operating advantage. AI sprinkled across everything at once, with no thesis about leverage, produces demos and press coverage but little durable value.
The moat is the data and the workflow, not the model
Founders sometimes worry that AI erases their advantage because everyone has access to the same models. That is precisely backwards. The model is the commodity. The advantage lives in the proprietary data you can feed it and the workflow you build around it — the parts competitors cannot copy by signing up for the same API.
A company that embeds AI into a workflow it uniquely understands, fed by data it uniquely holds, builds a moat that deepens with use. A company that bolts a generic chatbot onto its website builds nothing defensible at all.
Govern it before it governs you
The same power that makes AI valuable makes it a liability if it is ungoverned. A model making thousands of decisions is thousands of opportunities for a subtle, systematic error. Boards are right to ask how these systems are monitored, where a human stays in the loop, and how the company would know if something went wrong. This is not bureaucracy; it is the difference between AI as an asset and AI as an unpriced risk.
The companies that will look prescient in a few years are not the ones talking most about AI today. They are the ones quietly rebuilding a few core processes around it, protecting the data that makes it work, and governing it like the serious infrastructure it has become.