The Ninety Percent Was Never About Technology
Seventy five percent of boards approved major AI investment this year. Forty eight percent of those same boards have not set governance expectations for what they approved.
That gap is not an oversight. It is the boardroom telling you, plainly, what it believes AI actually is: a technology purchase, reviewed like one, funded like one, and expected to perform like one the moment it is switched on.
I have spent nine issues of this newsletter finding the same problem in nine different rooms of the house.
In May, it was memory: large language models operating without governed context, reasoning with no institutional recall, mistaken for intelligence when they were only ever fluent. In June, it was energy: the digital ambition of an enterprise running headlong into the physical limits of a power grid nobody priced into the model. It was spend discipline, watching organizations mistake usage volume for productivity and pay for the confusion twice. It was leadership signal, where I argued that a workforce is rarely failing to adapt to AI; it is failing to receive a coherent instruction from the people above it. It was the political consent nobody priced into a data center's site selection, lineage, ownership, uptime mistaken for trustworthiness, and finally the realization that AI governance is not a new discipline at all, just data governance wearing a more expensive name.
Nine domains. One diagnosis, repeated with different vocabulary each time: the interface gets funded. The foundation gets deferred. Every time.
This is not a technology failure pattern. It is an organizational one, and it recurs because it is comfortable. Funding the visible layer feels like progress. It produces a demo, a dashboard, a rollout announcement. Funding the invisible layer, governance, lineage, definitional ownership, produces nothing you can put in a slide until the year it prevents a failure nobody outside the room will ever hear about. Boards fund what they can see. Ninety percent of what determines whether an initiative survives contact with production is exactly what they cannot.
Three things I would do differently if I were sitting where that 48 percent is sitting right now.
First, stop treating the governance backlog as compliance overhead and start pricing it as capital allocation. Grant Thornton's research found a tenfold difference in audit pass rates between organizations with fully integrated governance and organizations still piloting it. That is not a risk statistic. That is a return statistic, and it belongs in the same conversation as the technology budget, not a footnote beneath it.
Second, stop waiting for a regulatory trigger that has already arrived. Texas's own Responsible AI Governance Act took effect January 1, 2026. Organizations that had already done the governance work met that date as a formality. Everyone else is now operating under a live statute they are retrofitting compliance onto after the fact. That is the more expensive way to arrive at the same place.
Third, audit where trust actually lives in your organization before you scale autonomous execution any further. My doctoral research found that executive sponsorship and governance functioning as a trust mechanism predict a data initiative's return more reliably than which technology stack sits underneath it. That finding has held in every engagement I have run. The technology question is rarely the real question. The trust question is.
None of this is a new argument. It is the same one I have made nine times since May, from nine different angles, because the pattern does not change industries or use cases. It changes costumes.
You cannot deploy what you have not governed. You cannot govern what you built without discipline in the first place. Everything else, the interface, the model, the demo that impressed the board, is the ten percent sitting on top of a decision that was already made, correctly or not, long before anyone asked what AI could do for the business.
The foundation was never invisible because it did not matter. It was invisible because nobody was pricing it. That has to change first.
If you found this briefing valuable, share it with a colleague who is navigating the shift from AI hype to operational reality.
