AI Doesn't Fix an Operating Model. It Inherits One.

The easiest part of enterprise AI adoption may be buying the licenses.

The harder question is what happens next.

Give a capable team access to increasingly capable models and they will find ways to use them. Content gets created faster. Analysis accelerates. Research expands. Workflows that once required hours can happen in minutes.

That can create real value… It can also accelerate problems the organization hasn't resolved.

Automation inherits the logic beneath it

Every automated process contains assumptions.

• What information is authoritative?

• What constitutes an exception?

• Who can approve an action?

• Which definitions should be used?

• When does a human need to intervene?

• Who is accountable when the output is wrong?

Traditional software often forces organizations to confront some of these questions during implementation.

Generative AI can make it surprisingly easy to avoid them.

A team can begin using AI long before the organization has established the operating rules surrounding that use.

The technology works.

The operating model hasn't caught up.

Speed changes the economics of ambiguity

Before AI, an unclear process might create a handful of inconsistent decisions each week.

Automate it, and the same ambiguity can propagate across hundreds or thousands of actions.

The problem isn't that AI is unreliable by definition.

It's that intelligence is only as reliable as the operating reality beneath it.

Conflicting definitions don't disappear because a model can interpret natural language.

Broken ownership doesn't disappear because an agent can execute a workflow.

Poor information lineage doesn't become trustworthy because analysis happens faster.

At scale, ambiguity simply moves faster.

Governance does not mean stopping experimentation

The opposite response can be equally damaging.

Organizations see the risk and create so much centralized control that experimentation becomes impractical.

Every use case requires approval. Every output requires review. Teams become afraid to explore. The organization owns powerful technology that nobody can meaningfully use.

Good governance should make responsible experimentation easier, not harder.

That requires defining where teams have autonomy and where controls matter.

• Which tasks can AI assist?

• Which decisions can it recommend?

• Which actions can it execute?

• Where is human review required?

• Which information can models access?

• What evidence is necessary before an experimental workflow becomes part of normal operations?

Those are operating-model questions as much as technology questions.

Automate what you understand

I don't believe organizations need to perfect every process before using AI.

Waiting for perfect information is another way to avoid learning.

But there is a meaningful difference between using AI to explore a process and embedding AI into a process the organization doesn't understand.

Experimentation can help expose ambiguity.

Production automation can institutionalize it.

That's why I think the sequence matters:

1. Make it work.

2. Put it to work.

3. Scale what works.

Automation belongs after enough operational truth exists to know what “working” actually means.

AI can dramatically increase the speed at which organizations analyze, decide and act.

The competitive advantage won't come simply from moving faster.

It will come from being confident that what you're accelerating is worth scaling.