At Build 2026, Microsoft folded Fabric IQ into a broader Microsoft IQ system and introduced Rayfin, an open-source SDK and CLI that deploys agent-built applications straight into Fabric with their data routed to OneLake by default. The combination is Microsoft's answer to a problem that has crept up on every enterprise running agentic coding tools: the agent is cheap, but the dozen new backend databases each agent spins up to ship its app are not. Rayfin is positioned head-on against Supabase and Neon, and Microsoft IQ is the context plane those agent-built apps are supposed to read from and write back into.
The shadow microservice problem
The agentic coding boom has shifted the unit of data sprawl from the dataset to the application. A year ago, governance teams worried about shadow spreadsheets and one-off extracts. Now the same teams face shadow microservices, each with its own Postgres-compatible database, its own auth model, its own retention policy, and none of it visible to lineage tooling. VentureBeat's VB Pulse Q1 2026 tracker captured the underlying shift in a single number: hybrid retrieval intent among organizations with more than 100 employees tripled between January and March 2026, from 10.3 percent to 33.3 percent. Enterprises are no longer asking whether they need shared context. They are arguing about where it lives.
That is the gap Microsoft is trying to close. If every agent-generated application defaults to a governed backend that lands in the same lakehouse as the rest of the business, the silo problem stops compounding. If it does not, every Fortune 500 ends up with hundreds of small databases the platform team cannot audit or roll back.
Inside the four IQ surfaces
Microsoft IQ consolidates four context sources into one connection point. Work IQ captures how an organization operates day to day, drawing from email, documents, meetings, and schedules. Foundry IQ curates institutional knowledge, indexing knowledge bases, rules, and procedures. Fabric IQ models the live operational state of the business through entities, relationships, and rules grounded in Fabric Real-Time Intelligence signals, with ontologies expected to reach general availability in the coming months. Web IQ adds real-time global context from the open web.
The integration story matters as much as the surface count. Wiring a single agent into operational data, document stores, a vector index, and a web tool typically means four connectors, four credential sets, and four freshness contracts. Microsoft IQ collapses that into one. Amir Netz, CTO of Microsoft Fabric, framed the goal bluntly: "Our job in the world of data is creating reality for agents based on data."
Why Rayfin targets Supabase and Neon
Rayfin is the more consequential announcement for data engineering teams. It is an open-source SDK and CLI that deploys agent-built applications directly to Fabric, routes their data to OneLake by default, and feeds that data back into Microsoft IQ. Netz described the architecture as bidirectional: the agent draws from the organization's ontology, and the data the application generates then enriches that ontology for the next agent. The flywheel only works if developers actually choose the governed path.
The competitive target is explicit. Microsoft named Supabase and Neon, the two Postgres-compatible backends that agentic coding tools most often pick by default. The argument is not that either product is bad. It is that a Supabase instance spun up by an agent is invisible to the rest of the enterprise data layer the moment it goes live. Multiply that by several hundred agent-generated apps across a large bank or retailer, and the silo problem rebuilds itself faster than any governance team can chase it. Rayfin's bet is that if the governed backend is the default deployment target in the CLI, the silo never gets created in the first place.
Robert Kramer of KramerERP framed the open question well: does Microsoft simplify execution and strengthen trust, or does it add another layer to an already complex environment. The answer turns on Rayfin's ergonomics. If spinning up a Rayfin app feels heavier than spinning up a Neon database, agents will route around it through whatever path their coding tool defaults to. If it feels lighter, the architecture wins by inertia.
Where the multi-cloud gap shows up
Microsoft IQ is a Fabric story, and that is its main limitation. Organizations on Snowflake, Databricks, or BigQuery will need an equivalent context plane, and the most credible interoperability vector is Open Semantic Interchange. The competitive frame is clarifying. Snowflake announced Horizon Context and Cortex Sense the same week. Redis is shipping Iris. Pinecone is shipping Nexus. Databricks is consolidating Iceberg v3 with unified governance. Every major data platform now has a context layer story, and the difference is no longer whether it exists but where it lives.
Microsoft's distinguishing bet is that the deployment target for agent-built apps and the context layer are the same surface. That removes a class of integration work that Snowflake and Databricks customers will still have to do themselves, at least until Iceberg-native equivalents catch up. Whether that lead is worth a re-platform is a separate question, and for most organizations the answer is no.
Our Fabric pilot plan for Q1 2026
If we are already standardized on Fabric, the right move is a single-tool Rayfin pilot before the end of Q1 2026. We would pick one internal agentic coding surface, scope it to two or three apps under a combined 50,000 dollar budget, and measure what fraction of the resulting app footprint lands in OneLake versus an external Postgres. The metric is governance coverage, not feature parity. Anything below 80 percent in the first quarter means Rayfin's CLI is not the path of least resistance yet, and we should escalate to the Fabric account team rather than absorb the silo cost quietly.
If we are not on Fabric, we treat this announcement as a forcing function. By end of Q2 2026, we need a written internal standard that names a default governed backend for agent-built apps on our primary cloud, whether that is a Snowflake-fronted target, a Databricks app deployment pattern, or a sanctioned Neon configuration with lineage hooks. The worst outcome is letting agentic coding tools choose the path of least resistance for us, because that path is almost always a new database the platform team cannot see.
The near-term test is concrete. Fabric IQ ontologies are expected to hit GA within the next two quarters, and Rayfin's adoption will be measurable by how many third-party agentic coding tools name it as a default deployment target by 30 September 2026. If that number is under five, the governed-default thesis is in trouble and Supabase keeps winning by inertia. If it clears ten, every other hyperscaler will be forced to ship an equivalent before the end of the year.
