There is a pattern emerging in enterprise AI that I think is underappreciated.
Everyone is building AI apps. Vibe-coded, prompt-engineered, agent-powered apps. The demos are impressive. The production deployments are sparse.
The companies actually getting AI into the hands of thousands of employees are not doing it through standalone AI apps. They are doing it through software those employees already use every day.
CRM. ERP. ITSM. The productivity suite. The integration layer.
AI is being diffused through the software estate — not bolted on top of it.
Why This Matters
When AI is embedded into software employees already use, you solve three problems simultaneously:
The adoption problem. You don't have to convince anyone to use a new tool. They are already in the tool. The AI is just there, contextually, when they need it.
The context problem. The software already has the context — the customer record, the open ticket, the transaction history, the workflow state. The AI doesn't have to retrieve it. It's already there.
The trust problem. Employees trust the systems they've used for years. They're appropriately skeptical of new AI interfaces. Embedding AI into trusted systems inherits that trust.
The Integration Imperative
This is why integration infrastructure matters more — not less — in an AI-first world.
The data that makes AI contextually useful — the customer history, the product usage, the support interaction — lives across dozens of systems. Getting it to the AI, at the moment the AI needs it, without breaking governance or latency requirements, is an integration problem.
Every enterprise that has underinvested in its integration estate is now paying a compounding penalty. They don't just have a data silo problem. They have an AI readiness problem.
What I'm Seeing
The enterprises moving fastest on AI are not the ones with the most sophisticated AI capabilities. They are the ones with the cleanest, most connected integration layers.
They built the pipes before they knew what would flow through them.
The enterprises moving slowest are spending most of their AI budget on data remediation — trying to clean and connect the underlying estate before they can do anything interesting on top of it.
The Practical Implication
If you are thinking about enterprise AI strategy in 2026, the question isn't just "what AI capability do we buy?" It's "what is the state of the underlying infrastructure that will deliver that AI to the people who need it?"
AI will be diffused through software. The quality of the software estate — and the integration layer connecting it — will determine how much AI value organizations actually realize.
The picks-and-shovels play in enterprise AI is not the AI itself. It's the infrastructure that makes AI contextually useful at the point of work.
Views are personal.