Microsoft's Build answer to AI data silos: a unified context layer and a governed app backend

Microsoft IQ and Rayfin aim to stop every new AI agent from starting from scratch — but the execution is still unproven.

Written by Lena Armitage · Bureau Tech · Jun 19, 2026

The problem every new agent recreates

Every AI agent an enterprise deploys starts with the same deficit: no memory of how the organization works, where its data lives, or what rules apply. That's a solvable problem when you have one agent. It becomes structural when agentic coding tools are spinning up applications faster than any governance team can track them — each one a potential new data silo sitting outside the platforms that are supposed to keep data coherent.

Microsoft used Build 2026 to announce two products aimed at both halves of that problem: Microsoft IQ, a unified context layer for agents, and Rayfin, a governed deployment backend for agent-built applications.

What Microsoft IQ actually is

Microsoft IQ is an expansion of Fabric IQ, an existing business data context layer, into a broader four-source system. The goal is a single integration point that gives any agent access to:

- **Work IQ** — organizational context drawn from email, documents, meetings, and schedules - **Foundry IQ** — institutional knowledge, including rules, procedures, and curated knowledge bases - **Fabric IQ** — live operational state of the business, modeled through data entities, relationships, and business rules via Fabric Real-Time Intelligence - **Web IQ** — real-time global signals from the web

The ontology layer inside Fabric IQ — the part that captures operational context — is not yet generally available. Microsoft says GA is expected in the coming months, which means the most technically novel piece of the system is also the least proven in production.

Amir Netz, CTO of Microsoft Fabric, described the platform's role using a *Matrix* analogy: the cascading code wasn't atmosphere, it was the layer that built the world the agents operated in. "Our job in the world of data is creating reality for agents based on data," he told VentureBeat.

What Rayfin does differently

RAG — retrieval-augmented generation, the technique of grounding model responses in retrieved documents — doesn't solve the application proliferation problem. When an agent builds a new app, that app needs a backend. Without a governed path, it defaults to whatever the coding tool reaches for, typically a Postgres-compatible service like Supabase or Neon.

Rayfin is Microsoft's answer: an open-source SDK and CLI that deploys agent-built applications directly to Fabric, landing their data in Microsoft OneLake by default. Netz described the relationship as bidirectional — an agent building a Rayfin application draws from the organization's ontology, and the data that application generates feeds back into that ontology for the next agent.

The differentiator Microsoft is claiming is governance: one data and compliance layer for the entire application fleet, rather than a growing collection of isolated backends.

A crowded field with no clear winner

Microsoft is not building this category alone. Snowflake announced semantic context capabilities this week. Pinecone has its Nexus platform, positioning its vector database as a broader knowledge engine. Redis has developed Iris, its context and memory platform.

The convergence is notable. VentureBeat's VB Pulse Q1 2026 RAG Infrastructure Market Tracker found that hybrid retrieval intent among organizations with 100 or more employees tripled from 10.3% in January to 33.3% in March — a sign that enterprises have moved past the question of whether to use AI and are now focused on the infrastructure underneath it.

Robert Kramer, managing partner at KramerERP, put the stakes plainly: "The real question is whether Microsoft simplifies execution and strengthens trust or adds another layer to an already complex environment."

That question doesn't have an answer yet. The ontology layer isn't GA. Rayfin is new. The market data shows demand for exactly what Microsoft is building — but demand and delivery are different things.

Key takeaways

Hybrid retrieval intent among 100-plus employee organizations tripled from 10.3% to 33.3% between January and March 2026, per VentureBeat's VB Pulse Q1 RAG Infrastructure Market Tracker — a signal that enterprise AI architecture concerns have shifted from model access to data plumbing.
Microsoft IQ consolidates four previously separate context sources — Work IQ, Foundry IQ, Fabric IQ, and Web IQ — into a single integration point for AI agents, covering organizational workflows, institutional knowledge, live operational data, and real-time web signals.
Rayfin, a new open-source SDK and CLI, deploys agent-built applications directly to Microsoft Fabric, routing application data into OneLake by default rather than creating isolated Postgres-compatible backends outside the governed data layer.
Microsoft is not alone: Snowflake, Pinecone, and Redis all announced or expanded shared context capabilities this week, confirming the category is competitive and the winner is not yet clear.
Fabric IQ's ontology layer — which captures live operational context — is not yet generally available; Microsoft says GA is expected in the coming months.

FAQ

What is Microsoft IQ and how is it different from Fabric IQ?

Fabric IQ was Microsoft's existing layer for modeling live business data context. Microsoft IQ expands that into a four-source system, adding Work IQ (organizational workflows), Foundry IQ (institutional knowledge and rules), and Web IQ (real-time web signals) alongside the original Fabric IQ operational data layer. The intent is a single integration point so a developer connecting a new agent gets all four context sources at once.

What is Rayfin and why does Microsoft think it matters?

Rayfin is an open-source SDK and CLI that routes agent-built applications directly to Microsoft Fabric, storing their data in OneLake by default. Microsoft positions it against Postgres-compatible backends like Supabase and Neon, which agentic coding tools often default to. The governance argument is that Rayfin keeps application data inside the same compliance and data layer as the rest of the organization, rather than creating isolated silos.

What is RAG, and why isn't it enough?

RAG stands for retrieval-augmented generation — a technique that grounds a model's responses in documents or data retrieved at query time, rather than relying solely on what the model learned during training. It helps with accuracy and currency, but it doesn't solve the problem of agents lacking shared organizational context, and it doesn't govern where application data lands when agents build new software.

Is Microsoft the only company building shared context infrastructure for AI agents?

No. Snowflake, Pinecone, and Redis all announced or expanded related capabilities around the same time. The convergence suggests the category is real, but it also means Microsoft's execution — not its concept — will determine whether Microsoft IQ and Rayfin gain traction.

What parts of Microsoft IQ are not yet available?

The ontology layer inside Fabric IQ, which captures live operational context, has not reached general availability. Microsoft says GA is expected in the coming months. That's the most technically distinctive component of the system, so its production readiness is worth watching.

Citations

Enterprise AI agents keep creating data silos. Microsoft's Build answer is Microsoft IQ and Rayfin.Hybrid retrieval intent among 100-plus employee organizations tripled from 10.3% in January to 33.3% in March 2026, per VentureBeat's VB Pulse Q1 2026 RAG Infrastructure Market Tracker.
Enterprise AI agents keep creating data silos. Microsoft's Build answer is Microsoft IQ and Rayfin.Microsoft IQ unifies four context sources — Work IQ, Foundry IQ, Fabric IQ, and Web IQ — into a single agent integration point.
Enterprise AI agents keep creating data silos. Microsoft's Build answer is Microsoft IQ and Rayfin.Rayfin is an open-source SDK and CLI that deploys agent-built applications directly to Fabric, routing data into OneLake by default.
Enterprise AI agents keep creating data silos. Microsoft's Build answer is Microsoft IQ and Rayfin.Robert Kramer, managing partner at KramerERP, said the real question is whether Microsoft simplifies execution and strengthens trust or adds another layer to an already complex environment.