The model is the engine. It is not the moat.

We are obsessing over the wrong layer.

A new model lands and the conversations start straight away. This one is great. That one is almost as good. This model is cheaper. That model is better at reasoning. This one has a bigger context window.

Those conversations matter. But I think we are obsessing over the wrong layer.

Here is the uncomfortable part. If you can access that model, your competitor probably can too. If you can buy that copilot, they can buy it. If you can use that agent framework, API or platform, they can probably use it too.

So if your AI strategy is basically “we have access to the best model”, how long is that really an advantage? A month? Six months? Maybe a year?

Think about cars. If every manufacturer suddenly got access to the same engine, would every car become the same? Of course not. The engine is only one part of the vehicle. The transmission matters. The controls matter. The navigation matters. The safety systems matter. The driver matters.

AI is heading in exactly the same direction. The model is the engine, and models are becoming increasingly accessible. The much more interesting question for leaders is: what are you building around the model that your competitors cannot easily copy?

That is where I believe the real AI moat starts. It is also what I call the Evolved Enterprise. The companies that win with AI will not necessarily have the best model. They will have built the best system around it.

AI activity is not AI capability.

So where does the moat actually start? One of the best ways to answer that is with a simple question: if your three best AI users left the company tomorrow, what AI capability would remain?

A lot of organisations today have plenty of AI activity. People are prompting. Teams are experimenting. Copilots are being rolled out. Pilots are happening everywhere. Someone has built an impressive agent. That is progress. But activity is not the same as capability.

If all of that knowledge lives inside a few enthusiastic employees, the organisation has not really built anything yet. AI capability is what remains when the individual leaves.

It means the organisation has started turning experimentation into something repeatable: shared ways of working, trusted context, defined workflows, clear governance, measurable outcomes, and knowledge that other people can reuse.

That is the shift leadership teams need to pay attention to. Not “how many people are using AI?” but “what can our organisation now do reliably that it could not do before?” The moat does not start when people use AI. It starts when AI capability becomes organisational capability.

One embedded workflow beats ten pilots.

I would take one AI workflow properly embedded into the business over ten disconnected AI pilots. A pilot proves that AI can do something. A workflow changes how the organisation actually works.

Imagine an employee has an important client meeting tomorrow. In one organisation, they open an AI tool and type “help me prepare for this client”. They copy some information in, get an answer, and maybe it helps.

Now imagine another organisation where the meeting itself triggers a workflow. The system:

• retrieves the approved client context
• identifies what has changed
• analyses relevant information
• checks the applicable rules
• prepares a briefing
• flags anything uncertain
• sends it to the employee for review

And after the meeting, the employee’s corrections and the outcome are captured, so the next interaction starts with better context.

Same AI model. Completely different organisational capability. That is the shift that matters: from AI sitting beside the work to AI becoming part of how the work gets done.

Once AI is embedded into a workflow, you start accumulating something far more valuable than prompts. You accumulate context, decisions, exceptions, human judgement, feedback and organisational learning. The real asset is no longer “we use AI”. It becomes “we have encoded how our organisation performs this job exceptionally well”. That is much harder to copy.

Context, then control, then intelligence.

Now comes the part that usually looks complicated: enterprise AI architecture. Models. Agents. RAG. APIs. MCPs. Vector databases. Memory. Orchestration. It can get technical very quickly. But at leadership level, you can simplify most of it into three layers.

Enterprise Context

This is what the organisation knows. Data. Documents. Business definitions. Customer history. Relationships. Policies. Institutional knowledge. Without this layer, AI is mostly operating with generic intelligence.

The Control Plane

This determines what is trusted and permitted. Permissions. Identity. Governance. Evaluation. Security. Human approvals. Boundaries. It is the layer that answers: what can the AI access, what can it do, and when must a human step in?

Intelligence

This is the layer everyone gets excited about. Models. Agents. Reasoning. Planning. Generation. Action.

Here is the mistake I see organisations making: they start with Layer 3. “Let’s build agents.” “Let’s automate this.” “Let’s connect the model to everything.” But a powerful agent sitting on weak context and weak controls does not create a powerful enterprise. It creates risk at scale.

The architecture should really be Context → Control → Intelligence, because agents are only as reliable as the context and controls underneath them. Strong intelligence on weak foundations is still weak architecture.

Anyone can buy the engine.

Put it together and look at what is happening. Models are getting better, cheaper, faster and more accessible. Your competitor can increasingly access the same intelligence you can. So the differentiator cannot simply be “we use the best model”.

The differentiator becomes everything you build around it. The context you own. The data you have accumulated. The workflows you have designed. The systems you have connected. The judgement you have captured. The governance you operate within. The feedback loops you generate. And probably the most important one: the speed at which your organisation learns.

That is why I keep coming back to the car. The AI model is the engine. But an enterprise also needs:

• The fuel: proprietary context and data.
• The transmission: workflows.
• The road connections: systems and APIs.
• The sensors: the agentic loop.
• The navigation: goals and instructions.
• The driver: people and critical judgement.
• The road rules: governance and guardrails.

None of these things alone creates the moat. The moat is how they work together. And underneath all of it, the architecture still simplifies into the same three layers. Enterprise Context asks what the organisation knows. The Control Plane asks what is trusted, permitted and governed. Intelligence asks what can reason, decide and act.

When these start operating as one system, AI stops being something employees occasionally open in a separate window. It becomes embedded in how the organisation senses, understands, decides, acts and learns. Again and again.

That is what I mean by an Evolved Enterprise. Not a company with more AI. A company that has built an organisational system around intelligence.

And the final point is the same one I started with. Anyone can buy the engine. Almost nobody can copy the entire car. That is the moat.

Which part of this architecture is your organisation missing today?

This essay brings together the five posts of The AI Moat Series, first published on LinkedIn on 24 to 26 August 2026.

1. Post 1: The model is the engine, not the moat
2. Post 2: AI activity is not AI capability
3. Post 3: One embedded workflow beats ten pilots
4. Post 4: Context, Control, Intelligence
5. Post 5: Putting the whole thing together

The views and ideas in this essay are my own and do not necessarily represent those of my employer.