Enterprise

Microsoft bets Build 2026 on agentic AI with Scout, seven MAI models, and Microsoft IQ

At its San Francisco developer conference, Microsoft unveiled an always-on Autopilot agent for Microsoft 365, a new in-house reasoning model, GPU-accelerated Fabric queries, and a unified context layer aimed at the data-silo problem agents have created.

By Asher Glick · Enterprise software desk · 2026-06-04 · 4 min read

Photo: Unsplash / Teemu Paananen — Audience seated facing a brightly lit developer conference keynote stage

Microsoft used the June 2 opening of Build 2026 to introduce Scout, an always-on Autopilot agent for Microsoft 365 built on the open-source OpenClaw framework, and to surround it with the infrastructure stack the company now considers prerequisite to agentic computing: a new in-house reasoning model, GPU-accelerated data warehousing, and a unified context layer called Microsoft IQ.

Scout reaches into Teams, Outlook, OneDrive, and SharePoint through Work IQ, the context engine that supplies it with a user’s documents and patterns. Access is gated: enrollment in Microsoft’s Frontier early-adopter program and a paid GitHub Copilot subscription are both required. Every agent runs under its own governed Entra identity rather than a shared service account, which is the part of the announcement that matters most for any IT department that has spent the last two years watching ungoverned bots accumulate permissions.

“We all have our interesting quirks in how we work, and people are codifying those patterns into memories and skills that persist in their agent,” said Omar Shahine, Scout VP. The framing is deliberate. Microsoft is selling persistence, not autonomy.

The Microsoft AI Superintelligence Team also released seven in-house models, headlined by MAI-Thinking-1, the company’s first reasoning model. Microsoft says it was trained from scratch on commercially licensed data, with no distillation, a claim that reads as a direct pitch to enterprise legal departments still nervous about training-data provenance. The MAI lineup will be distributed through Fireworks AI, Baseten, and Open Router, with Fireworks now generally available on Azure AI Foundry.

On the data side, Fabric Data Warehouse now runs eligible queries directly on NVIDIA accelerated computing inside the execution engine. Microsoft’s May 2026 internal benchmarks claim up to 7x faster performance than three comparable cloud warehouses at 64-user concurrency, a vendor-marked number that’ll get audited the moment customers run it themselves.

The closing move is Microsoft IQ and Rayfin, both pitched as the answer to a problem Microsoft itself helped create: AI agents generating their own data silos and ungoverned front-ends. Microsoft IQ extends Fabric IQ with three new context sources; Rayfin wires agent-written front-ends back into Fabric’s data and identity layer.

It’s the familiar Microsoft pattern. Sell the agents, then sell the governance plane required to contain them.

Sources