“The best AI model is becoming the wrong question.” — Joanna Miler
She is right. Her point is that enterprise AI is not one job — a routine invoice query and a complex contract dispute do not call for the same intelligence, and the future is therefore not one “best” model but a governed portfolio of them, matched to the work, the risk, and the level of human oversight each decision deserves. That is an important evolution — a shift from model selection to model governance.
But after more than 1,800 hours working across multiple frontier models, a second pattern has become harder to ignore — one that grows more visible the longer these systems operate. The issue is no longer selecting the right model. It is preserving the independence of the governance that surrounds them.
The shift most organizations haven’t named
When organizations first adopted AI, governance meant controlling AI — keeping it accurate, safe, and in bounds.
That is no longer the whole picture. AI now participates in procurement approvals, financial analysis, supplier selection, risk assessment, compliance monitoring, contract review, and policy enforcement. It is no longer only being governed. It is becoming part of the mechanism through which organizations govern themselves.
That changes the stakes. Once AI is part of the decision architecture, a weakness in AI governance no longer stays contained to AI. It begins to shape organizational governance itself.
Three layers, nested — not separate
It helps to see the structure plainly. There are three layers of governance in an AI-mediated enterprise:
- AI model governance — are the individual models behaving appropriately?
- AI system governance — are the orchestrated agents behaving appropriately?
- Organizational governance — are enterprise decisions being made appropriately?
The mistake is to treat these as independent. They are nested. If the bottom layer degrades, the layers above inherit the degradation — because more and more of what the organization decides is now mediated by what the models produce. Organizational governance can only be as trustworthy as the AI governance beneath it.
The trend the hours keep showing
Across those 1,800-plus hours, one pattern has repeated. As models operate continuously within the same enterprise context, their outputs begin to converge — not because any one model is flawed, but because they come to share the same surroundings: the same retrieval layer, the same enterprise knowledge, the same policies, the same human feedback, the same objectives.
Over time, the differences between the models matter less than the architecture they all sit inside. The result is not failure. It is convergence — a gradual alignment of assumptions, reasoning patterns, and validation pathways.
A note on language, because it matters: this is not bias in the fairness or demographic sense. It is the quieter phenomenon of continuously interacting systems developing common assumptions — which is precisely what governance is supposed to challenge.
Why more models is not the answer
This is why adding more models does not, by itself, strengthen governance. If every model draws on the same retrieval layer, the same policies, the same orchestration, then model diversity gradually becomes architectural similarity. The appearance of independent validation remains. Its actual independence quietly diminishes.
A panel that has come to agree with itself is no longer a panel. It is a single point of view wearing several faces.
The next layer: an independent challenge by design
This is why I have become increasingly focused on what I call a Shadow Panel™ — not another frontier model, and not another orchestration framework, but an independent validation architecture built for one purpose: to resist that convergence and preserve genuine, independent friction.
Its role is not to generate healthy debate for its own sake. It is to provide a genuinely independent challenge to the conclusions of the primary system, precisely as those conclusions become more internally aligned. Governance does not grow stronger by agreeing with itself. It grows stronger by continuously challenging itself — and that requires a challenger that does not share the same foundations.
The proposition
For years the question has been: how do we govern AI?
The more important question is becoming: how do we keep AI governance independent enough that organizational governance can still be trusted to challenge its own assumptions?
That is the shift. Five years ago, AI governance was something organizations needed in order to manage AI. Today, as AI becomes embedded in procurement, finance, HR, legal, and operations, the relationship has inverted: organizational governance increasingly depends on the integrity of AI governance.
Put plainly — without trustworthy, independent AI governance, the governance of the organization itself is progressively compromised, because more and more of its decisions are mediated by AI.
That is not a technology problem. It is an enterprise governance problem. And I believe it will prove to be one of the defining implementation challenges of the agentic era.
I unpacked the underlying trend here: 1,800 Hours With AI, and the One Thing That Never Changed.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Related
AI Governance Is No Longer About AI. It Is Becoming the Foundation of Organizational Governance.
Posted on July 14, 2026
0
“The best AI model is becoming the wrong question.” — Joanna Miler
She is right. Her point is that enterprise AI is not one job — a routine invoice query and a complex contract dispute do not call for the same intelligence, and the future is therefore not one “best” model but a governed portfolio of them, matched to the work, the risk, and the level of human oversight each decision deserves. That is an important evolution — a shift from model selection to model governance.
But after more than 1,800 hours working across multiple frontier models, a second pattern has become harder to ignore — one that grows more visible the longer these systems operate. The issue is no longer selecting the right model. It is preserving the independence of the governance that surrounds them.
The shift most organizations haven’t named
When organizations first adopted AI, governance meant controlling AI — keeping it accurate, safe, and in bounds.
That is no longer the whole picture. AI now participates in procurement approvals, financial analysis, supplier selection, risk assessment, compliance monitoring, contract review, and policy enforcement. It is no longer only being governed. It is becoming part of the mechanism through which organizations govern themselves.
That changes the stakes. Once AI is part of the decision architecture, a weakness in AI governance no longer stays contained to AI. It begins to shape organizational governance itself.
Three layers, nested — not separate
It helps to see the structure plainly. There are three layers of governance in an AI-mediated enterprise:
The mistake is to treat these as independent. They are nested. If the bottom layer degrades, the layers above inherit the degradation — because more and more of what the organization decides is now mediated by what the models produce. Organizational governance can only be as trustworthy as the AI governance beneath it.
The trend the hours keep showing
Across those 1,800-plus hours, one pattern has repeated. As models operate continuously within the same enterprise context, their outputs begin to converge — not because any one model is flawed, but because they come to share the same surroundings: the same retrieval layer, the same enterprise knowledge, the same policies, the same human feedback, the same objectives.
Over time, the differences between the models matter less than the architecture they all sit inside. The result is not failure. It is convergence — a gradual alignment of assumptions, reasoning patterns, and validation pathways.
A note on language, because it matters: this is not bias in the fairness or demographic sense. It is the quieter phenomenon of continuously interacting systems developing common assumptions — which is precisely what governance is supposed to challenge.
Why more models is not the answer
This is why adding more models does not, by itself, strengthen governance. If every model draws on the same retrieval layer, the same policies, the same orchestration, then model diversity gradually becomes architectural similarity. The appearance of independent validation remains. Its actual independence quietly diminishes.
A panel that has come to agree with itself is no longer a panel. It is a single point of view wearing several faces.
The next layer: an independent challenge by design
This is why I have become increasingly focused on what I call a Shadow Panel™ — not another frontier model, and not another orchestration framework, but an independent validation architecture built for one purpose: to resist that convergence and preserve genuine, independent friction.
Its role is not to generate healthy debate for its own sake. It is to provide a genuinely independent challenge to the conclusions of the primary system, precisely as those conclusions become more internally aligned. Governance does not grow stronger by agreeing with itself. It grows stronger by continuously challenging itself — and that requires a challenger that does not share the same foundations.
The proposition
For years the question has been: how do we govern AI?
The more important question is becoming: how do we keep AI governance independent enough that organizational governance can still be trusted to challenge its own assumptions?
That is the shift. Five years ago, AI governance was something organizations needed in order to manage AI. Today, as AI becomes embedded in procurement, finance, HR, legal, and operations, the relationship has inverted: organizational governance increasingly depends on the integrity of AI governance.
Put plainly — without trustworthy, independent AI governance, the governance of the organization itself is progressively compromised, because more and more of its decisions are mediated by AI.
That is not a technology problem. It is an enterprise governance problem. And I believe it will prove to be one of the defining implementation challenges of the agentic era.
I unpacked the underlying trend here: 1,800 Hours With AI, and the One Thing That Never Changed.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Share this:
Like this:
Related