Enterprise AI agents continue to create data silos. The Microsoft Build response is Microsoft IQ and Rayfin.

Every new artificial smart agent deployed by your team starts from scratch: do not remember how the business works, where the data is located or what rules apply. Since the proxy coding tool has started its applications faster than anyone manages them, each tool has the potential to become a completely isolated island outside the data layer. Microsoft These two issues will be addressed directly on Build 2026.

According to the first season of 2026 of the VB Pulse of Venture Beat degrees RAG The infrastructure market tracking report, with mixed search intentions in over 100 employee organizations increasing from 10.3 per cent in January to 33.3 per cent in March, is a threefold increase, indicating that enterprises are no longer expanding RAG coverage and are now focusing on their bottom structures. Sharing business context is an unsolved part of the search.

In terms of context, Microsoft is expanding its existing business data context layer, Fabric IQ, into a broader unified system, known as Microsoft IQ, with three additional context sources, covering the way the organization works, what it knows and real-time global signals from the network, so that any agent can use all four context sources as a basis. Rayfin is a new open source for applications. SDK and CLI, which deploys proxy-built applications directly to Fabric as a managed back end of production, directs the application data to the same platform instead of creating a new island.

Microsoft Fabric Chief Technical Officer Amir Netz explains the scope of the data platform in film metaphors. The green screen of the hacker empire's middle-link code is not the atmosphere, but the layer of the world that Agent Smith operates.

Netz told Venture Beat: "Our work in the data world is to create reality for data-based agents."

Microsoft IQ Aligns four context sources to a proxy base Medium

Microsoft IQ brings together the four context sources that have existed separately so far, so that developers can connect new agents to all four context sources through a single integration step.

Work IQ. Captures the organization ' s day-to-day mode of operation, using e-mail, documentation, meetings and scheduling to inform the seating table about people, teams and workflows.

Castery IQ. Management of the knowledge, planning and indexing knowledge base of the agency to enable the agent to understand the meaning of working within the organization, the applicable rules and the procedures to be followed.

The fabric IQ. Modelling the real-time operational status of operations through data, defining entities, relationships and operational rules based on real-time signals from Fabric real-time intelligence. This is expected to reach GA version in the coming months.

Internet IQ. Add a real-time global context from the network to provide proxy with the current situation in the outside world of the organization and its internal data.

“The agent will be a well-informed virtual employee”, Netz said. “That is the direction of the world.”

Rayfin links proxy-built applications to the same data base

Build shared context solves half of the problem. The other is what happens when the agent starts generating the application. Each new application requires a backend, and without a controlled deployment path, each application will create a new data island entirely outside the context layer.

Rayfin provides an enterprise backend and deploys proxy-built applications directly to Fabric, so the application data defaults to Microsoft OneLake and feeds back to Microsoft IQ context layers rather than to their external accumulation.

Microsoft locates Rayfin in Supabase and Neuon, which is a backend compatible with Postgress, which is used by proxy encoding. The difference is in governance: Rayfin uses the Fabric unified data and compliance route to the whole application group, rather than creating isolated islands.

Netz describes this relationship as a two-way relationship. The agent who built the Rayfin application drew on the organization ' s own body. The data generated by the application will then enrich the body of the next agent.

Every major data platform is pursuing the same answer, but the results of implementation are not proven

Microsoft is not the only platform to build a shared context layer for the agent. Snowflake announced her context function and semantic function this week. Pinecone has the Nexus platform that expands the vector database to a knowledge engine, and Redis has developed the Iris context and memory platform.

Microsoft's approach further reinforces the trend that RAG and model availability are no longer a problem.

The KramerERP management partner Robert Kramer told Venture Beat: “Fabric IQ and Rayfin are important because the challenges of enterprise artificial intelligence are no longer related only to model availability.” “The real question is whether Microsoft will simplify implementation and enhance trust or add another layer to an already complex environment.”