In the past few weeks the AI hype has shifted from context graphs (and yes, everyone thinks slightly differently about what they actually are and can deliver) to the AI tech stack lasagne.

Some big vendors, like OpenAI are making plays for various layers of the stack - stretching out from model layer to the orchestration layer. While this market continues to heat up - we will see more and more major vendors trying to own as much of the stack as they can. While this might seem convenient, this is usually not good news for enterprises (as history shows with mediocre monoliths), and the impact will quickly and acutely be felt in loss of agility and getting stuck in vendor bound logic.

The success of AI for enterprises hinges on excelling in both flexibility and control - which is best achieved by leveraging horizontal layers of best in class technology. Layers designed to work across vendor agnostic stack components, that evolve independently and that do not collapse in to one roadmap.

At IndyKite, we believe control should increase as systems become more autonomous. But not coarse-grained, static, after-the-fact control. Control needs to be dynamic and intelligent, operating at speed at the moment data is retrieved and used.

That changes what enterprises can do.

They can experiment with different models and orchestration layers without rebuilding governance every time. They can introduce new agents without fragmenting data access and use policies. They can evolve parts of the stack while keeping a consistent layer of enforceable, traceable control across it.

Context is central to this. Not just to improve agent decisions, but to ensure that data carries its provenance, consent, and usage constraints wherever it moves. Context becomes the mechanism that makes governance enforceable at runtime, across the ecosystem.

The stack will evolve quickly. Enterprise architecture needs to remain free and flexible enough to evolve with it.