Your AI strategy died in the org chart

Your AI strategy didn’t fail because the models weren’t ready. It failed the day someone routed it to the IT department.

I understand the reflex. For thirty years, anything with a login screen went to the CIO. ERP, CRM, the data warehouse, the collaboration suite. That routing rule worked because those were systems to be installed. You bought them, integrated them, trained people on them, and the business kept operating the way it always had, just with better plumbing.

AI is not plumbing. AI is a decision about how work gets done, which makes it an operating decision, which means it belongs to the people who own operations. Route it to IT and you haven’t started an AI initiative. You’ve started a procurement exercise wearing an AI costume.

Technology theater

Here is what the IT-owned version looks like, and you’ll recognize it because it is probably happening in your building right now.

A steering committee forms. A governance framework gets drafted. Three vendors are invited to a bake-off. A pilot launches in a “low-risk” corner of the business, chosen precisely because nothing important happens there. Six months later there is a chatbot on the intranet that summarizes the HR policy manual, a slide deck about responsible AI, and a roadmap of proofs of concept stretching into next year.

Everyone is busy. Budgets are consumed. And not one number the CEO stares at has moved.

I call this technology theater, and I don’t say it to be cruel. Every person in that process is doing their actual job well. The CIO is managing risk, which is what CIOs are paid to do. IT is measured on uptime, security, and cost control, so IT rationally optimizes for containment. But AI’s value doesn’t live in containment. It lives in variance, in doing the work differently than you did it last quarter. Asking the department whose job is preventing surprises to lead your most surprising capability is an org-chart error, not a talent problem.

What the other version looks like

Business-owned AI is quieter and much less photogenic. There is no council. There is a P&L owner who picked one process she controls, mapped every step of it, and asked a blunt question about each step: is this judgment, or is this pattern?

Judgment stays with people. Pattern gets rebuilt around the machine. Then she reprices the process, restaffs it, and moves the savings or the speed into whatever she was starved for. The AI part is almost boring. The redesign is the whole game.

I know this version works because I live inside it. I run a company where I am the entire engineering department. Using agentic systems, I ship what used to take a team of twenty, while also carrying sales and strategy, because those were mine before the machines showed up and they’re mine now. When I rebuilt our production pipeline at Vidloft, the hard part was never the models. The hard part was admitting how much of our workflow existed because of constraints that no longer applied, and having the nerve to delete it.

Notice what I didn’t do. I didn’t become IT. I spent two decades as an operator, a founder, a producer. I made ten thousand videos, directed a documentary, sold to Fortune 500 marketing departments. I went deep on the technology because I refused to let it stay abstract, not because I wanted to change professions. That’s the posture I’m arguing for. Not “executives should learn to code.” Executives should refuse to accept any description of AI that doesn’t mention their own workflows by name.

The Monday morning test

Here’s a test you can run this week without hiring anyone.

Ask for the AI roadmap. Whoever sends it to you, note who that was, because you just learned who owns it. Then read the document and ask one question. Does this look like our operating plan, or does it look like an IT project list?

An operating plan names processes and owners. It says “quote-to-cash cycle, currently eleven days, target four, owned by Marcus.” An IT project list names platforms and phases. It says “Phase 2: enterprise LLM gateway, Q3, pending security review.” Both documents can be competently written. Only one of them can make you money.

If you want a second test, try this one. If every AI pilot in your company vanished tonight, which P&L owner would notice by Friday? If the honest answer is none of them, you don’t have an AI strategy. You have an AI department.

And if the roadmap fails both tests, the fix is not a better roadmap. The fix is a different owner. Hand the question to the executive who feels it when margin slips, give them real authority over how their function runs, and let IT do what IT is genuinely great at: keeping the thing secure, reliable, and sane once the business has decided what the thing is for.

The leverage was never in the stack

The uncomfortable part, the part I’d want a friend to tell me, is that the stack question is mostly settled and mostly rented. The models are converging. The tooling is a credit card away. Your competitors can buy every piece of technology you can buy, at the same price, this afternoon. There is no durable advantage sitting inside a vendor contract.

What can’t be bought is the decision. Which processes get rebuilt first. Which roles change shape. Which customers feel it, and how fast. Those are strategy questions, and strategy questions have never once in the history of the org chart been answered by procurement.

I’ve watched this from both sides. I’ve been the vendor in the bake-off, sitting across from a committee that had no authority to change how anyone actually worked. And I’ve been the operator with my hands inside my own P&L, rebuilding a workflow at midnight because I finally understood which half of it was pattern. One of those rooms produces decks. The other one produces margin.

The companies pulling ahead right now aren’t the ones with the best AI. They’re the ones where an operator with real authority looked at their own work, the actual daily mechanics of it, and started making calls only an owner can make. The technology was the easy part. It usually is.

Your org chart is a routing table. Right now it may be routing the most important operating question of the decade to the one department that was explicitly designed not to answer it. Change the route, and the stack takes care of itself.