The first era of enterprise AI focused heavily on infrastructure.

GPUs. Cloud platforms. AI factories. Foundation models.

The next wave brought Generative AI into everyday work through copilots, assistants and increasingly capable AI agents.

Each layer has expanded what organisations can compute, understand and automate.

But a fundamental enterprise problem remains:
How does all this intelligence become a decision the organisation can actually rely on?

That question becomes increasingly important as AI moves beyond generating content and starts influencing pricing, risk, investment, operations, resource allocation, compliance and other consequential business decisions.

The infrastructure exists.
The models exist.
The intelligence exists.

What has been missing is the Decision Layer.

Generative AI has transformed how organisations work with information.

It can generate:

• emails
• reports
• summaries
• analysis
• recommendations
• knowledge answers
• plans
• possible courses of action

These capabilities are enormously useful.

But an answer is not yet an organisational decision.

Consider the decisions enterprises make every day:

Should we approve this transaction?
What price should we offer this customer?
Where should scarce capacity be allocated?
How much inventory should we replenish?
Which investment should receive priority?
Should this exception be approved or escalated?

These decisions require more than generating a plausible answer.

They depend on the organisation's actual:

• objectives
• business realities
• alternatives
• constraints
• policies
• evidence
• trade-offs
• authority

And importantly, organisations need to know:
Why was this decision made?

Most organisations already have ways of making important decisions.

But those ways of deciding are often fragmented.

Part of the decision may exist in policy.
Part may exist in an ERP or CRM.
Part may exist in spreadsheets.
Part may live in the experience of senior employees.
Part may emerge during meetings.

And increasingly, part may be delegated to AI systems and agents.

This becomes a structural problem as AI adoption scales.

If every team, application or AI agent reconstructs the decision independently, the organisation can end up with:

different interpretations
different reasoning
different priorities
different outcomes

from essentially the same situation.

The missing layer is therefore not simply another AI model.

It is a way of making the organisation's model of the decision explicit.

Large Language Models are extraordinarily capable at working with language.

They can interpret requests, synthesise information, generate explanations and reason over increasingly complex problems.

But language describes organisational reality.

It is not the organisational reality itself.

A consequential enterprise decision may depend on a customer, transaction, product, supplier, asset, employee, obligation, policy, available resource or changing operational condition.

These things exist in relationships with one another.

And the decision may depend on which relationships matter, which objectives apply, which constraints cannot be violated and which trade-offs the organisation is willing to make.

That is why the next enterprise AI architecture requires something different:
Enterprise Decision Models.

An Enterprise Decision Model makes explicit how an organisation determines what should be done for a particular class of decision.

It can represent:

Business Reality
What is actually happening?

Objectives
What is the organisation trying to achieve?

Alternatives
What can be done?

Constraints
What limits cannot be violated?

Policies and Rules
What organisational requirements apply?

Evidence
What information supports the decision?

Trade-offs
When objectives compete, what matters more?

Authority
Who can decide, approve, challenge or override?

The organisation's way of deciding becomes something that can be inspected rather than remaining implicit inside prompts, meetings, documents or individual judgement.

They can use tools, access systems, coordinate tasks and execute increasingly complex workflows.

But the ability to act creates an even greater need to establish how consequential decisions should be made.

A workflow can determine: What happens next?

An agent can determine: What action can I take?

But the organisation still needs to establish:
What should be done, under these conditions, according to our objectives, constraints, policies and authority?

Otherwise, scaling agents can also mean scaling different interpretations of how the organisation should decide.

The Decision Layer provides a shared organisational foundation against which humans, AI systems, agents and workflows can operate.

Enterprise AI has largely been model-centric:
Which LLM should we use?
Which model performs best?
Which agent framework should we deploy?

Those questions remain relevant.

But organisations also need to become decision-model-centric.

Every organisation has its own:

• customers
• products
• resources
• objectives
• policies
• risk appetite
• constraints
• priorities
• authority structures

These realities determine how the organisation should make decisions.

They should not have to be reconstructed from scratch every time someone asks an AI system a question.

Instead, they can become part of an explicit organisational Decision Model.
The AI model can change. The organisation's Decision Model remains under organisational control.

That distinction becomes increasingly important as models, vendors, capabilities and AI architectures continue to evolve.

The emerging enterprise architecture can therefore be understood as three complementary layers.

1. Infrastructure Layer
Provides the compute required to run modern AI. 

2. Intelligence Layer
Language models, predictive models, analytics and other AI systems generate, interpret and analyse intelligence.

3. Decision Layer
Enterprise Decision Models determine how organisational realities, objectives, alternatives, constraints, evidence and authority should be evaluated to determine what should be done.

These layers do not compete.

They complement one another.

Infrastructure provides compute.
AI provides intelligence.
The Decision Layer determines what should be done.

The question for the C-suite is gradually changing.

It is no longer simply:
Where can we deploy AI?

It is becoming:
Which important decisions are being influenced by AI, and how does our organisation want those decisions to be made?

That creates a different set of questions.

Can we explain why the decision was reached?
Can we see which constraints mattered?
Can we determine which evidence was used?
Can we identify the trade-offs?
Can we reproduce the decision?
Can we govern who has authority?
Can we change the underlying AI model without unintentionally changing the organisation's way of deciding?

These are not merely AI-model questions.

They are organisational decision questions.

Organisations need help identifying:
which decisions matter
how those decisions are currently made
which realities and constraints determine them
where AI should contribute intelligence
where human judgement and authority must remain
how the decision should be governed

This creates a different transformation opportunity.

Instead of merely implementing another AI tool, organisations can make their existing decision-making explicit and improve it systematically.

The result is not simply another AI application.

It is an organisational capability.

Governments want AI that is useful, safe, explainable and accountable.

But as AI becomes increasingly involved in consequential decisions, governing the AI itself is only part of the problem.

Institutions must also be able to govern:
how decisions are made
which policies apply
which constraints cannot be violated
what evidence is required
who has authority
when human review is necessary

This is where AI governance begins to meet decision governance.

AI models will continue becoming more capable.

Agents will become more autonomous.

AI will become embedded across more organisational processes.

But greater intelligence does not remove the need for organisations to decide.

It makes the way organisations decide more important.

The next era of enterprise AI will therefore not be defined only by larger models, more agents or more compute.

It will also be defined by whether organisations can make their important decisions:
explicit
explainable
consistent
governed
recomputable as reality changes

That is the role of the Decision Layer.

And it is the foundation of what we now call DecisionAI.