Most conversations about AI start from the technical side. Because that is the easy entry point. There is a rubric. Models, benchmarks, capabilities. Something to point at. But what if we start somewhere else.
What is a company?
Not in the legal sense. Not the org chart or the mission statement. I mean functionally. What is the thing we are actually trying to augment with AI? Here is how I think about it.
A company is a chain of decisions that converts resources into something a customer will pay for.
That is it. Follow any product to market and you find a long sequence of decision nodes behind it. Do we build this? Who supplies the components? How much do we make? What do we charge? Which customers do we prioritize? Link enough of those nodes together and you have what we call a firm. It sounds reductive. But I think it is the right level of abstraction.
Because once you see a company as a chain of decisions, something else becomes visible. Its entire operational history is a record of past decisions. Approved vendors. Rejected loans. Campaigns funded or killed. Claims paid or denied. Tickets escalated or closed. Every entry has the same underlying structure.
Context → Decision → Outcome.
Anyone who has worked with machine learning will recognize that immediately. It is exactly what you need to train any ML / AI system. Which means most companies are not sitting on a data problem. They are sitting on years of decision data they have never thought to use that way. Now here is where I think the framing gets genuinely useful.
AI is not intelligence.
That word carries too much baggage and leads people in the wrong direction. Toward science fiction, toward replacement narratives, toward arguments about consciousness. A more precise description:
AI is the ability to codify decision-making.
That is all it is. And that framing matters. Because humans have always codified decisions. Arithmetic. Accounting rules. Credit scoring models. Statistical process control. We have been building decision machinery for centuries, even longer, should I run or should I fight?
What has changed dramatically with modern AI is the range of what type of decisions we can codify. Historically, we could only formalize decisions with clean inputs and explicit rules. The moment a decision required reading context, interpreting ambiguity, or pattern-matching across messy unstructured information — it stayed in human hands. Not because we didn’t want to automate it. Because we couldn’t. Now that looks different Pattern recognition. Document interpretation. Language reasoning. Judgment calls that previously required a trained human and a salary to match, all of those decision points are now within reach.
The set of codifiable decisions is expanding. The cost of executing them is collapsing.
So what is the question?
From this viewpoint, companies as decision chains, and AI as codified decision making the key question is not where can we apply AI? But: which decisions, currently trapped in human workflows, can now be formalized, and what happens to the business if we can run them at scale?
That reframe opens up a much wider aperture. Because organizations are not short on decision problems. They are running thousands of them every day, manually, at significant cost in time and headcount. Fraud signals evaluated one by one. Invoices touched by human hands before they route. Support tickets read and sorted by someone whose judgment is probably consistent but whose attention is finite. Documents reviewed. Leads scored. Risk assessed. None of it glamorous. All of it plentiful.
And plentiful is the key word. The economics of AI do not work on rare, high-stakes decisions where every case is genuinely novel. They work where volume is high, structure is present, and the cost of running one more decision approaches zero.
The companies that figure this out first are not asking which AI tool to buy. They are auditing their decision chains — finding the nodes that are high volume, under-formalized, and sitting on years of outcome data — and asking what it would mean to run those at ten times the current throughput.
Shift your mental model
So here is the mental model I would hand to any executive trying to cut through the AI hype.
You are not running a company. You are running a decision factory.
Every day, your organization makes thousands of decisions. Most of them invisible. Most of them never examined as a class of problem. They just happen distributed across departments, absorbed into salaries, treated as the natural cost of doing business. That assumption is now worth questioning. Because if AI is codified decision-making, and the marginal cost of executing a codified decision is approaching zero, then the relevant question is not which AI product to buy. It is: where in my decision factory does volume matter?
Not every decision qualifies. Novel decisions with genuinely unique context still require human judgment. High-stakes decisions with significant downside still need accountability structures around them. It is between those those extremes your fertile middle ground sits. This is where you have real leverage.
So the next time someone in the boardroom asks what your AI strategy is, do not lead with tools or vendors or transformation roadmaps. Come in with a map of your decision chain.
Here is where we make a thousand decisions a week by hand. Here is what that costs us in time, headcount, and missed volume. Here is where the logic is consistent enough to formalize. And here is what the business looks like if we run those decisions at ten times the current throughput.
That is an AI strategy. Everything else is theater.