Microsoft pitches Windows and Azure as the full-stack agent platform at Build 2026

At a two-day developer conference in San Francisco, the company introduced an open agent governance spec, OS-level execution controls, a PostgreSQL-compatible agent database, and Project Solara — its bet on agent-first hardware.

At Build 2026 in San Francisco, Microsoft spent two days and a two-and-a-half-hour keynote making one argument: that Windows and Azure together constitute the platform on which AI agents will be permissioned, isolated, billed, and eventually embodied. The product slate maps neatly onto the classical stack, governance, runtime, data, hardware, and the strategic point is that Microsoft intends to own every layer.

The centerpiece is the Agent Control Specification, an open governance standard that defines permissions, human-approval triggers, logging, and interception points for agent behavior. An ACS SDK ships with plug-ins for LangChain, the OpenAI Agents SDK, the Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI, and MCP tools. Publishing a spec that wraps your competitors’ SDKs is itself a strategy. It mirrors the EMV-style playbook of past platform incumbents: define the rails, let everyone ride them, set the defaults.

Underneath ACS, the security and OS layer hardened. Agent 365 SDK went generally available with built-in observability, access controls, and compliance enforcement. The Microsoft Execution Container SDK brings OS-level isolation for agent processes on Windows, and Windows 365 for Agents, a fully isolated, policy-governed Cloud PC built to run any agent, also hit GA. Agents now have a sandbox with a SKU.

On data, Azure HorizonDB entered public preview as a fully managed, PostgreSQL-compatible database. Microsoft says its GPU-accelerated Fabric Data Warehouse, built on custom CUDA kernels developed with NVIDIA, ran 7x faster than three comparable external vendors at 64-user concurrency in internal benchmarks in May 2026. NVIDIA’s Ian Buck framed the work as “low-latency performance for many simultaneous users.”

Then came the hardware turn. Satya Nadella introduced Project Solara, a platform for agent-first devices with screens and microphones that host agents interfacing with cloud systems rather than running conventional apps. “Whenever these new platforms come, you get to rewrite even the rules of how new platforms operate,” Nadella said. NVIDIA’s Jensen Huang and Qualcomm’s Cristiano Amon appeared on stage in support.

Microsoft also released MAI Thinking-1, its first reasoning model, which it claims matches Anthropic’s Claude Opus 4.6, days after Anthropic shipped Opus 4.8. A new Copilot agent called Scout, for email and message triage, was previewed. The frontier-model race continues; the platform race is what Microsoft thinks it can win.

Microsoft pitches Windows and Azure as the full-stack agent platform at Build 2026

Sources