Your Enterprise Doesn't Need Another AI Platform. It Needs an Intelligence Layer.

My last piece on enterprise silos, “Your AI is working. So Where is the ROI”, generated more conversation than I expected.

People reached out directly—technology executives, vendors, and others working on enterprise AI. Different industries, different perspectives, but many of them recognized the same problem: AI may be advancing extraordinarily quickly, but the enterprises trying to deploy it are still fragmented across decades of systems, data, processes and organizational boundaries.

It made me think I had hit a nerve.

And it raised a more interesting question.

If silos are the problem, what exactly are we supposed to do about them?

For 30 years, enterprise technology has tried to eliminate silos.

It hasn't worked.

AI may finally force us to stop trying.

The Enterprise Isn't Going to Be Rebuilt

Every generation of enterprise technology seems to arrive with a promise of simplification.

ERP would consolidate the enterprise.

Data warehouses would consolidate the data.

Cloud would consolidate the infrastructure.

Data lakes would consolidate the data again.

Thirty years later, the enterprise is arguably more fragmented than ever.

And now AI platforms promise another layer of consolidation.

Yet walk into almost any large enterprise today and the reality looks very different.

Oracle remains Oracle. Salesforce remains Salesforce. ServiceNow remains ServiceNow. Databricks remains Databricks. Mainframes remain mainframes. AWS, Azure and GCP remain where workloads run.

It's time we stop pretending we're going to eliminate the silos.

Because many of those silos exist for perfectly rational reasons.

A trading platform wasn't designed to work like a CRM system. A payments platform has different requirements from a data lake. A risk system has different controls from a marketing application. Companies acquire other companies. Regulations change. Technologies evolve. Business units make different decisions.

Complex enterprises produce complex technology estates.

The question isn't how to make all of that complexity disappear.

The question is how AI can understand it.

AI Changes the Problem

Traditional enterprise integration has largely been about moving information.

Get the customer data from system A into system B.

Build an API.

Create an ETL pipeline.

Put everything into a warehouse or lake.

Synchronize another database.

Those approaches remain necessary. But AI introduces a different requirement.

AI doesn't simply need access to information.

It needs context.

This one is worth repeating.

AI needs Context.

Which customer is this?

Which system is authoritative?

What does this field actually mean?

What other systems are related to this transaction?

Who is permitted to see this information?

What happened before?

What business rule applies?

What action is this person authorized to take?

And increasingly:

What is the AI itself allowed to do?

The more we move from copilots that answer questions to agents that actually perform work, the more important these questions become.

An agent that can reason but cannot understand the enterprise around it isn't particularly useful.

Another Platform Isn't the Answer

This is where I think enterprise AI architecture may be heading in the wrong direction.

Every technology wave produces platforms. AI is producing a lot of them.

Enterprises are being asked to choose AI platforms, agent platforms, model platforms, data platforms, vector databases, orchestration frameworks and governance platforms.

Many of these technologies are excellent.

But there is a danger that, in trying to solve the enterprise silo problem, we simply create the newest silo—the AI silo.

Imagine an AI platform with fantastic models, sophisticated agents and beautiful interfaces.

Then ask it to resolve a real customer problem.

Suddenly it needs information from Salesforce, transaction history from Oracle, a document sitting in SharePoint, an incident from ServiceNow, telemetry from AWS and an authorization rule buried somewhere else.

The model isn't the problem.

The enterprise around it is.

The Intelligence Layer

Perhaps what enterprises need isn't another destination for their data.

They need an intelligence layer across the systems they already have.

Not something that replaces Oracle, Salesforce, ServiceNow, Databricks, AWS or the mainframe.

Something that understands how they relate.

Such a layer would need to understand far more than data. It would need to understand relationships, semantics, identity, permissions, policies, lineage and business processes.

It would need to know not simply where information lives, but what that information means in the context of the enterprise.

And it would need to expose that understanding safely to models and agents.

That is fundamentally different from another repository.

The data can remain where it belongs.

The applications can remain where they belong.

The infrastructure can remain where it belongs.

What changes is our ability to reason across them.

From Integration to Understanding

For decades, enterprise architecture has largely connected systems so that they could exchange information.

AI raises the bar.

Now the enterprise needs to become understandable—not just to people, but to machines.

That may ultimately be one of the most important architectural changes created by enterprise AI.

The winning architecture may not be the one that finally eliminates the silos.

We have been trying to do that for 30 years.

It may be the one that finally makes the silos intelligible to each other.