2026-06-08 — views

Microsoft Build 2026: 'Microsoft IQ' makes context the platform — Work IQ, Foundry IQ and a new Web IQ go live

Read this because We skipped the obvious MAI-models headline because the durable Build 2026 story is structural: Microsoft wants grounded context to be a default platform layer, not a feature. Watch Web IQ — a low-token web-grounding broker.

At Build 2026 Microsoft folded grounding into one brand, Microsoft IQ: Work IQ, Foundry IQ, Fabric IQ and a new Web IQ — context as managed infrastructure.

What Microsoft actually announced

The headline from Microsoft Build 2026 was a set of in-house MAI models. But the more consequential platform move was quieter: Microsoft pulled its grounding and retrieval tools under one umbrella called Microsoft IQ, and declared it generally available across GitHub Copilot, Microsoft Foundry and Copilot Studio.

The framing in Microsoft’s own keynote post was blunt: context becomes critical infrastructure. The argument is that enterprise agents rarely fail because the model is weak; they fail because the agent cannot see the right document, the right meeting, or the right live web fact at the moment it reasons. Microsoft IQ is the answer to that gap, and it has four layers.

The four layers

Work IQ is the workplace-intelligence layer. It captures how work happens across Microsoft 365 — people, emails, documents, meetings and how they connect — and exposes it to agents. The public Work IQ APIs (REST, plus A2A and MCP protocols) reach general availability on June 16, 2026.

Foundry IQ is the knowledge layer for agents, and it is generally available now. Its knowledge bases unify Work IQ, Fabric IQ, File Search, Azure SQL and MCP sources behind a single, SLA-backed retrieval endpoint. Microsoft reports the agentic retrieval improves recall by up to 54 percent versus single-shot RAG and lifts answer quality by up to 20 percent across evaluated datasets, while spending fewer tokens. A serverless Developer tier is in public preview.

Fabric IQ Ontology is the business-semantics layer — a shared model of how people, data, workflows and operations relate, so agents query real financial and operational data instead of guessing. It is in preview.

Web IQ is the genuinely new piece: an AI-native grounding API that discovers, ranks, extracts and packages fresh information from web pages, news, images and video. Microsoft cites 164ms P95 latency, roughly 2.5 times faster than the next best alternative, with fewer tokens per query. It is in limited access via a waitlist.

Layer	Job	Status
Work IQ	Microsoft 365 work signals	APIs GA June 16, 2026
Foundry IQ	Unified enterprise knowledge	GA (Serverless in preview)
Fabric IQ	Business semantics / ontology	Preview
Web IQ	Live web grounding	Limited access (waitlist)

Why this matters

For two years the AI conversation has been a model race. Build 2026 signals a different fight: whoever owns the context pipe owns the agent. A model is increasingly a commodity you can swap; the data plumbing that decides what the model sees is sticky, permissioned, and tied to where your work already lives. By branding all of it Microsoft IQ and wiring it into Copilot, Foundry and Copilot Studio at once, Microsoft is trying to make grounding a default rather than a project.

The retrieval numbers are the part to scrutinize. A 54 percent recall lift and 20 percent answer-quality gain are real engineering claims, but they are vendor benchmarks on undisclosed evaluated datasets. Recall improvements also cut both ways: surfacing more candidate context can raise answer quality and raise the surface area for leaking the wrong document if permission and sensitivity-label enforcement is imperfect. Microsoft is explicit that Foundry IQ carries permissions sync and Purview sensitivity-label auditing in preview — which is reassuring precisely because it concedes the risk.

Practitioner note

If you build agents, the actionable shift is to stop hand-rolling retrieval. Foundry IQ’s single SLA-backed endpoint over Work IQ, Fabric IQ, Azure SQL and MCP is meant to delete the custom plumbing most teams maintain today. Pilot it against your existing RAG stack and measure the recall and token claims on your own data before you trust the 54 percent figure — and verify that sensitivity labels and permission sync behave under adversarial queries, not just happy-path demos. Treat the GA dates as gates: Foundry IQ is usable now, Work IQ APIs land June 16, and Web IQ is waitlist-only, so design fallbacks for the layers you cannot yet call.

The under-considered angle

Web IQ is the sleeper. Everyone is reading Microsoft IQ as an enterprise-data play, but a fast, model-agnostic, low-token live-web grounding API is also a lever over the open web’s economics. If agents increasingly read the web through a Microsoft passage-retrieval layer rather than crawling and rendering pages themselves, Microsoft sits between publishers and the agents consuming their content — shaping what gets ranked, extracted and cited, and how few tokens (and how little attribution) flow back. The model race is loud; the quiet contest over who brokers context may decide more.