Microsoft Build 2026: Enterprise Agents Become a Data Context Race

Microsoft used Build 2026 on 2 June to argue that the next phase of enterprise AI agents will be decided less by model capability and more by the data context those agents can reach. The headline announcements were Azure HorizonDB, Fabric IQ general availability, GPU-accelerated Fabric Data Warehouse, and the Work IQ APIs, all aimed at the same problem: agents that start cold every time and cannot ground decisions in organisational knowledge.

For enterprise buyers and suppliers, the signal is clear. The agent stack is widening again, and the value is moving towards the layer that gives agents shared semantic context across the business.

What was actually announced

Azure HorizonDB is a fully managed PostgreSQL service positioned for agentic workloads. Microsoft is claiming up to three times faster transactions and search performance compared with self-managed PostgreSQL, with built-in vector indexing, semantic search, in-database access to AI models, and integrations with Microsoft Fabric and Foundry. The pitch is straightforward: keep PostgreSQL, add the AI-native database primitives that agents now need.

Fabric IQ has reached general availability as a relationship-first modelling engine that sits over structured business data. Microsoft framed it as a shared semantic foundation that lets multiple agents reason about the same business entities, relationships and metrics rather than each rebuilding its own view of the company. New Fabric IQ work includes an Operations agent, a Graph relationship model and updated grounding that combines internal context with external sources.

The Work IQ APIs, generally available from 16 June, expose Microsoft 365 signals as a programmable intelligence layer for agents. Web IQ adds grounded web access. Taken together, Microsoft is describing three intersecting context layers for enterprise agents: Work IQ over Microsoft 365 signals, Fabric IQ over structured business data, and Web IQ for external grounding.

Underneath, GPU-accelerated Fabric Data Warehouse, Foundry IQ as a connector layer to external data, and continued investment in the Foundry Agent Service round out the picture. Microsoft also reiterated its work with NVIDIA on Windows security primitives and OpenShell from last week's GTC Taipei announcements.

Why this matters for enterprise buyers

The most common reason enterprise agent pilots stall is not the model. It is that agents have no durable, governed view of the organisation. Each new agent rebuilds context, hits the same data permission walls, and produces inconsistent answers because the underlying entities, hierarchies and metrics are not shared.

Microsoft is now positioning data context as a buying decision. If Fabric IQ becomes the semantic layer for an organisation, every agent built on Foundry, Copilot Studio or third-party harnesses can inherit the same understanding of customers, products, orders and operations. HorizonDB extends that into transactional and vector workloads. Work IQ extends it into the day-to-day signals of how people actually work.

That changes the procurement conversation. Buyers should expect to evaluate not just the agent platform, but the context layer the agents draw on, who owns it, how it is governed, and whether non-Microsoft agents can use it on equal terms.

The widening agent stack

A recognisable enterprise agent stack is now emerging across the major platform vendors. There is a model layer, a harness, a runtime that enforces identity and policy, a skills layer, and now an explicit context layer that grounds agents in business data and signals.

NVIDIA's GTC Taipei announcements last week emphasised the runtime, harness and skills side through the Agent Toolkit, Nemotron 3 Ultra and OpenShell. Microsoft is now leaning hard on the context side. Google's Managed Agents work, AWS Bedrock agents and ServiceNow's AI Control Tower are pulling in similar directions. The practical implication for buyers is that no single vendor owns the whole stack, and architectural choices at each layer will shape what changes later.

Implications for the wider agentic AI market

Context layers are becoming a procurement category. Buyers should expect questions about semantic models, ontologies, graph data and grounding to appear in agent platform evaluations alongside the model and runtime.

Databases are becoming agent-aware. HorizonDB joins a broader trend of vector-native, AI-aware databases. Suppliers selling into the enterprise should be ready to explain how their agents integrate with these stores, not only with vector add-ons bolted on top.

Identity and access controls become more important, not less. If multiple agents share the same context layer, the controls on who and what can read each slice of that context become a central governance question.

What buyers should ask now

• Where is the durable context for our agents going to live, and who owns it?
• Can third-party agents and harnesses use the same context layer as the platform vendor's own agents?
• How are entities, metrics and relationships modelled, and how does that align with our existing data platform?
• What identity, permission and audit controls apply to context access, not just tool execution?
• How do we avoid lock-in at the context layer if we change agent platforms later?

The Agentic Expo angle

Agentic Expo exists for buyers and suppliers who are past the demo stage. Microsoft's Build 2026 announcements reinforce what is becoming the central question for production agents: can they reason over a shared, governed view of the business?

Suppliers that can describe how they plug into emerging context layers, how they respect identity and permissions on that data, and what evidence they have from real deployments will have a more productive conversation with the buyers walking the floor at Olympia in March 2027.