Essay · AI & Capital Discipline · Part 12 of 12
Ask a company what it's investing in and you'll get a polished answer. Ask what it has decided not to fund, and how recently, and you learn something real.
The best operators I know carry an actual list. Not a strategy document. A short, specific list of things they've decided they won't do, with reasons, that they revisit rather than rewrite.
Why a list beats a principle
Principles don't survive contact with a specific proposal. Everyone agrees not to fund vague initiatives, right up until the vague initiative arrives with an enthusiastic sponsor and a plausible slide.
A list is harder to argue around because it's concrete. "We don't fund pilots without a named production owner" is a rule the current proposal either passes or doesn't. The decision was made in the calm, and you're just applying it.
That's the real function. It moves the hard call to a moment when nobody's ego is in the room.
What tends to be on it
From the organizations I've watched do this well, some entries recur.
Initiatives whose only claimed benefit is time saved, with no downstream mechanism. Pilots with no production owner and no budget line waiting. Anything needing three functions to change behaviour simultaneously. Vendor commitments where the exit hasn't been priced. Any second-year renewal that can't state what it produced in year one. Anything requiring data to leave the boundary without a legal review that names the risk out loud.
None of these are absolute. Each is a strong default with an override that requires someone senior to put their name on it. That's the mechanism. Not prohibition. Friction, applied deliberately.
The list has to cost you something
A refusal list nobody has been annoyed by isn't working.
If everything you declined was obviously bad, you've written a document rather than a discipline. The list earns its keep the first time it blocks something a senior person really wanted, and you hold the line, and it turns out fine, and everybody notices that it turned out fine.
That's the moment the culture changes. Not when the list is written. When it's enforced against someone who could have overridden it.
And it has to be wrong sometimes
Any rule strong enough to be useful will block something that would have worked. That's not a flaw in the design, it's the price of the design, and pretending otherwise is how these things get quietly abandoned.
So put a review on the list itself. Once a year, look at what you refused and ask whether any of it was a mistake. Sometimes you'll find one. Change the rule, or don't, but do it deliberately rather than by exception.
Where this series lands
Eleven parts and the argument has been one thing throughout. The technology is not the variable. It never was.
What separates the organizations getting returns from the ones funding activity is a set of habits that would work in any cycle: attach a unit to the claim, name who owns it in production, design the stop as carefully as the start, scrutinize the renewal rather than the approval, sort decisions by what reversal costs, sequence for what each bet leaves behind, measure for the skeptic, and keep a list of what you won't do.
None of that is about AI. That's why it'll still be true for whatever comes next, and why the organizations that build it now will be the ones ready when it does.
The operating principle. You are defined by what you decline. If your refusal list has never cost you anything, you don't have one.
Juan Vegarra is the author of An Outsider’s Playbook. The views here are his own. More essays · Advisory · Write me