Skip to content

Microsoft Build 2026 — Fabric & Databases for Agentic Apps

Overview

At Microsoft Build 2026, Microsoft announced a set of data platform capabilities anchored on the thesis that the bottleneck for enterprise agentic AI is no longer model capability, but consistent, shared data context across the business. The announcements span a new PostgreSQL-compatible database (Azure HorizonDB), an open-source backend deployment SDK (Rayfin), a shared context layer for agents (Fabric IQ), GPU-accelerated analytics, and significant additions to Azure Cosmos DB and Azure SQL for agent tooling.

Azure HorizonDB (Public Preview)

Azure HorizonDB is a fully managed, PostgreSQL-compatible database designed specifically for AI-era workloads. It combines PostgreSQL familiarity with a cloud-scale distributed architecture.

Architecture and Scale

Dimension Specification
Compatibility PostgreSQL-compatible (open-source tools work as-is)
Storage Elastic up to 128 TB
Compute scale-out Up to 3,072 vCores
Commit latency Sub-millisecond, multi-zone
Availability Zone-resilient by default
Performance vs. self-managed PostgreSQL Up to 3x faster transactions and search

AI-Native Features

  • Hybrid vector + full-text search: Combines DiskANN (vector similarity via pgvector) with full-text search (pg_textsearch extension) in a single query — the pattern most RAG applications require. Simultaneously matches semantic meaning and keyword relevance.
  • In-database model invocation: The azure_ai extension brings model inference into the PostgreSQL engine via SQL, eliminating the external orchestration layer typical in current AI-database stacks.
  • Direct connectivity: Native integration with Microsoft Foundry and Microsoft Fabric; GitHub Copilot in VS Code support for accelerated development.

HorizonDB's Web IQ integration (see Microsoft IQ below) already underpins both Microsoft Copilot and ChatGPT's live web grounding.

Rayfin (Public Preview)

Rayfin is an open-source SDK and CLI that lets developers and coding agents define and deploy a complete application backend entirely in code, running directly on Microsoft Fabric.

How It Works

  1. Describe the app backend in code or natural language (to a coding agent).
  2. Rayfin generates a typed, governed backend: database schema, business logic, APIs, identity, and access policies.
  3. One CLI command ships it to Microsoft Fabric as a managed service.
  4. App data lands in OneLake by default — no copy, no ETL.
  5. The deployed backend inherits the tenant's existing security, governance, and compliance controls.

Key Integrations

  • Replit partnership: Build in Replit (natural language to running app), deploy with Rayfin — data and services remain in the organization's own Fabric tenant under existing identity and network controls.
  • Microsoft Fabric: Acts as a production application backend; every Rayfin app gets enterprise-grade security and scale from day one.

Problem Addressed

Moves agentic AI apps from prototype to production without teams having to build and manage backend infrastructure. Fabric becomes a viable application backend — not just a data warehouse.

Fabric IQ (Generally Available)

Fabric IQ is Microsoft's shared context layer for AI agents and real-time intelligence within Microsoft Fabric. It gives all agents operating within the Fabric ecosystem a single, consistent, governed view of the organization's data.

Components and GA Status

Component Status Description
Operations Agents GA (Build 2026) Autonomous agents for operational workflows on Fabric data
Graph GA (Build 2026) Relationship-first modeling engine for organizational knowledge
Planning GA (late June 2026) Enterprise planning capability: budgets, forecasts, scenario models on semantic models
MCP Server Preview Exposes Fabric IQ ontologies to any agent from any vendor via the Model Context Protocol

Architecture Role

Fabric IQ's business ontology is now accessible via MCP to agents from any vendor — not just Microsoft's. This positions Fabric IQ as a vendor-neutral shared context bus rather than a proprietary data silo.

Planning complements ontologies with integrated planning, giving agents a complete, contextual view of historical, real-time, and forward planning data from a single source.

GPU-Accelerated Fabric Data Warehouse

Query acceleration in Fabric Data Warehouse integrates NVIDIA accelerated computing to fundamentally change analytical query performance for AI-driven workloads.

  • Benchmark: Up to 7x faster than three comparable external vendors for reporting and application workloads at 64-user concurrency (internal benchmarking, May 2026).
  • Purpose: Designed for AI-driven analytical workloads where sub-second query response is required for agent reasoning loops.

Agent Skills for Power BI

  • Agent Skills for Power BI: Developers prompt an AI agent to build and refine semantic models and reports within Fabric.
  • Fabric Apps for Semantic Models: AI agents can build and deploy Fabric-native web apps on top of semantic models — moving from data exploration to deployed application without leaving the semantic layer.

Azure Cosmos DB Announcements

New Capabilities (Build 2026)

Feature Status Description
Cosmos DB MCP Toolkit Preview Connects Cosmos DB data to AI agents and Copilots via the Model Context Protocol
Semantic Reranking Preview Improves search relevance using built-in contextual understanding — no extra model calls required
Global Secondary Indexes Preview Enables efficient querying on non-partition-key attributes at global scale
Linux Emulator GA Build, test, and validate Cosmos DB apps locally on Linux, macOS, and Windows without a cloud dependency
Agent Memory Toolkit Preview Standardizes persistent memory for AI agents using Cosmos DB + Azure Durable Functions + Foundry models

The Agent Memory Toolkit addresses one of the most common production gaps in agentic systems: the absence of durable, portable memory that persists across sessions and agent invocations.

Azure SQL / SQL Server — Agentic Tooling

Feature Status Description
SQL MCP Server Public Preview Securely connects SQL data to AI agents and Copilots via Model Context Protocol
160 / 192 vCore options GA Larger instance sizes for high-throughput agent workloads
Vector index enhancements GA Faster, more capable vector indexes with improved search performance and efficiency — no code changes required

Microsoft IQ — Unified Intelligence Layer

Microsoft IQ is the umbrella name for Microsoft's unified intelligence layer, announced at Build 2026 as generally available. It unifies four intelligence streams:

Layer Description
Work IQ Intelligence grounded in Microsoft 365 organizational data
Foundry IQ Intelligence from models, knowledge bases, and agent execution data
Fabric IQ Intelligence grounded in organizational analytics and operational data
Web IQ Live web grounding for real-time external information

Foundry IQ knowledge bases unify Work IQ, Fabric IQ, File Search, Azure SQL, and MCP behind a single SLA-backed retrieval endpoint — simplifying multi-source retrieval for agent developers.

Microsoft Agent Framework 1.0 also reached GA at Build 2026. Copilot Credits is the named consumption metering unit for agent work across the Microsoft platform.

Best Practices

Challenge / Area Description Solution / Recommendation
Shared context across agents Multiple agents see inconsistent views of org data Use Fabric IQ as the shared ontology layer; expose via MCP for vendor-neutral access
RAG retrieval quality Vector search alone misses keyword-relevant results Use HorizonDB's hybrid DiskANN + pg_textsearch in a single query
Prototype-to-production gap Custom backend infrastructure delays agentic app deployment Use Rayfin to define backends in code and deploy to Fabric in one CLI command
Agent memory persistence Agent state lost between sessions Use Cosmos DB Agent Memory Toolkit (Cosmos DB + Durable Functions + Foundry models)
SQL data in agent tools Connecting relational data to agents requires custom integration Use SQL MCP Server (Public Preview) for governed MCP-based access
Analytics latency in agent loops Slow queries stall agent reasoning GPU-accelerated Fabric Data Warehouse targets sub-second latency for agent-driven analytics

See Also

References