Much of the confusion around AI today comes from the fact that very different kinds of work are being discussed as if they were the same thing.

They are not.

What is often described as “AI adoption” or “AI strategy” actually spans three distinct layers, each with its own incentives, pressures, and failure modes. Understanding the difference between them clarifies why so many conversations feel incomplete, and why certain questions never seem to get answered.

Layer One: Tool Creation

The first layer is tool creation.

This is the domain of vendors, developers, and platform builders: the people and companies creating models, applications, agents, features, integrations, and workflows.

Their incentives are clear and largely rational. They are under constant pressure to:

• ship new capabilities

• demonstrate progress and innovation

• respond to competitive threats

• expand use cases

• and support legacy implementations at the same time

Every new feature solves some problem.
Every release promises more capability.
Every roadmap implies momentum.

But tool creation carries an unavoidable cost: permanent obligation.

Once a capability exists, it must be maintained, documented, secured, supported, and defended. Technical debt accumulates. Backward compatibility matters. Expectations harden.

Tool creators rarely ask whether a feature should be used in a specific professional context. Their job is to make the feature possible.

That is not a criticism. It is simply the nature of the role.

Layer Two: Tool Operation

The second layer is tool operation.

This is the domain of consultants, implementers, IT teams, trainers, support providers, and internal champions: the people responsible for making tools work in real environments.

Operators live downstream from vendors. They absorb the churn.

They are under continuous pressure to:

• stay current on release notes

• understand changing interfaces and capabilities

• train staff on new functions

• troubleshoot failures and edge cases

• translate “what’s new” into “how to use this”

As tools evolve faster, this layer becomes more energized and more evangelistic.

New capabilities generate excitement.
Excitement drives adoption.
Adoption creates demand for training and support.

But this layer has its own blind spot.

Operators are rarely positioned, or incentivized to ask:

• Should this capability be used in this situation?

• Is this appropriate for this client?

• What obligation does this create?

• Who bears responsibility if this goes wrong?

They focus on how to use the tool, not whether it should be used.

Again, this is not a failure. It is a structural limitation of the role.

The Evangelism Loop

As the first layer accelerates innovation, the second layer amplifies it.

New features are demonstrated.
Use cases are promoted.
Success stories are shared.

Together, these layers create momentum that feels compelling and, at times, inevitable.

What gets lost in that momentum is deliberation.

The conversation shifts from “What are we deciding?” to “How quickly can we implement this?”

That is where risk quietly accumulates.

Layer Three: Judgment and Governance

The third layer is different.

It is judgment under uncertainty.

This layer is not about building tools or operating them. It is about deciding how a firm chooses to relate to AI at all given its responsibilities, obligations, and tolerance for risk.

This is where questions surface that the other layers cannot answer:

• Who is authorized to use AI, and for what?

• What kinds of work are off-limits?

• How do we reconcile capability with professional duty?

• What happens when guidance changes or conflicts?

• How do we respond when a client asks a question we didn’t anticipate?

These questions are not technical.
They are not operational.

They are governance questions.

AI governance lives above tools and operations, not as an obstacle, but as an orienting layer. It exists to ensure that decisions are made intentionally, rather than by default.

Why This Keeps Coming Up

Many firms feel as though they are having the same conversation about AI over and over again.

Everything seems fine.
Then something happens.

A new feature appears.
A client asks an unexpected question.
A regulator issues guidance.
An insurer changes language.
A colleague forwards an article that creates concern.

These are not constant events. They arrive irregularly, sometimes quietly, sometimes suddenly.

Between these moments, things appear stable. Then the ground shifts.

This is not because the firm is failing to keep up. It is because the environment itself is unstable.

AI Governance exists to handle exactly this condition.

What makes AI adoption feel so unsettled inside firms is not the technology itself. It’s the way conversations about it never quite resolve. Something new appears, a concern is addressed, a position is taken and then a few months later, a new feature, headline, client question, or industry signal reopens the discussion.

That pattern is not a failure of leadership. It’s the natural result of decisions with long-term responsibility being shaped by systems and roles designed for short-term capability.

Vendors build forward. Operators keep things running. Neither is positioned to pause and ask whether a particular capability belongs in this firm, for this work, under these obligations.

When that question goes unowned, it doesn’t disappear. It shows up as quiet drift, inconsistent assumptions, or a sense that the firm is always reacting slightly too late. Partners often realize what they believe only after a decision has already been acted on.

AI governance exists to hold that gap not as a rulebook, and not as an exercise in control, but as a way of staying oriented. So decisions are made deliberately rather than by default, and revisited when conditions change rather than avoided until they become urgent.

The goal is not to predict the future perfectly. It is to remain steady enough to recognize what is changing, what matters, and when attention is required.

__________________________________

Forwarded to you?  
Subscribe to receive future issues

 Subscribe