Shared by @futuretrees
AI's next enterprise lock-in is not the model. It is the context.
Chamath is right that token control matters. Routing work across OpenAI, Anthropic, Gemini, open-source models, small models, and whatever comes next gives companies leverage on price, latency, capability, and availability.
But token routing is only the first control plane. The deeper risk is letting the same vendor own the model, the agent layer, the workflow traces, the evals, the permissions, and the memory of how the company works.
Models will converge. Agent patterns will converge. Company memory will not.
A company's real context is made of customer promises, pricing exceptions, Slack debates, failed migrations, roadmap fights, support escalations, owner history, and decision scars. That does not live in model weights. It is not universal. It is yours.
Forward deployment makes this urgent. Embedded AI teams can create real value because they learn how the company actually operates. But that same mechanism can turn into dependency if the vendor's agents become the place where work is interpreted before it is acted on.
Query-time tool calls are useful, but they are not the same as durable organizational memory. Pulling from five systems when someone asks a question can feel like memory. It is not the same thing as owning a persistent, permissioned, inspectable context graph.
The right architecture is simple: rent intelligence from whoever is best, but own the context layer that makes intelligence useful.
The company brain should sit above source systems and below agents. It should know what happened, where the evidence lives, who is allowed to see it, what changed, what was promised, who owns it, and which ontology applies. Then any model can use that memory and any agent can act from it.
A decent model with great company context will beat a frontier model with shallow context in many enterprise workflows. In sales, support, engineering, and finance, knowing what actually happened matters more than sounding smart.
At Sentra, this is the layer we are building: a shared intelligence and memory system for the company, across communication channels, knowledge bases, actions, and agent traces.
Rent the intelligence. Own the context.