Microsoft Build 2026: The Final Day Delivered
Build 2026 wrapped in San Francisco with Day 2 delivering the announcements people will be talking about for months. If Day 1 established the foundation — Windows as an agent platform, RTX Spark hardware, Microsoft IQ as a unified intelligence layer — Day 2 built the house on top of it. Microsoft Scout. Majorana 2. In-house reasoning models. Frontier Tuning. And a new category of AI Microsoft calls Autopilots.
Here's everything that mattered across both days.
Day 1: The Foundation
Windows Becomes an Agent Platform
The keynote opened with Satya Nadella declaring that Windows is no longer just an OS for applications — it's a runtime for persistent AI agents. The [Windows Agent Framework](https://github.com/microsoft/agent-framework) went open source under MIT, providing four building blocks: an Agent Registration Service for lifecycle management, a declarative agent.json manifest, a gRPC-based cross-agent communication bus, and an encrypted Memory Service for conversational context.
The wagent CLI packages agents into deployable artifacts that run on Windows 11, Windows Server 2026, IoT, and Windows 365 Cloud PCs. Visual Studio 2026 ships with an Agent Designer and sandboxed local runtime. Microsoft even published a Linux version of the Agent Designer — a clear signal that agent development isn't locked to Windows even if Windows is the primary target.
RTX Spark: Hardware Built for Agents
Microsoft's partnership with NVIDIA produced RTX Spark-powered PCs delivering up to 1 petaflop of local AI performance, 6144 Blackwell RTX cores, and up to 128GB of unified memory. The Surface Laptop Ultra and Surface RTX Spark Dev Box were both announced — the latter a compact desktop with 128GB unified memory capable of running 120 billion parameter models locally. Windows implements workload profile scheduling optimized specifically for RTX Spark's 20-core architecture.
Microsoft IQ: The Intelligence Layer
Microsoft IQ launched as the umbrella for organizational intelligence, unifying four previously separate stories:
- •**Work IQ** — semantic understanding of emails, meetings, documents, and chats inside the Microsoft 365 trust boundary
- •**Foundry IQ** — knowledge bases unifying Work IQ, Fabric IQ, File Search, Azure SQL, and MCP behind one SLA-backed retrieval endpoint
- •**Fabric IQ Ontology** — shared business semantics defining how people, data, workflows, and operations relate
- •**Web IQ** — AI-native grounding APIs built on twenty years of Bing infrastructure, re-architected for LLMs and multi-step agents. 164ms P95 latency, highest grounding satisfaction, fewest tokens per query
Copilot Credits & Agent 365
Microsoft named the consumption meter: Copilot Credits. Agent work moves to pay-as-you-go ($0.01/credit or prepaid packs), while per-user Copilot stays for human Copilot use. Agent 365 expanded with an SDK (GA, free and framework-agnostic), Local Agents Public Preview (Claude Code, Copilot CLI, OpenClaw), and Windows 365 for Agents — Cloud PCs designed specifically for agent workloads.
Project Solara: An OS for AI Gadgets
Microsoft demonstrated Project Solara, a new OS built on Android (not Windows) designed specifically for AI agent gadgets. Two concept devices were shown: a desk-based smart display with facial recognition, and an AI ID badge. The goal is a platform purpose-built for ambient agent experiences rather than retrofitting agents onto traditional OS paradigms.
Day 2: The Headlines
Microsoft Scout: The First Autopilot
The biggest announcement of Build 2026 — and arguably the most significant product reveal — was Microsoft Scout. This isn't another Copilot. It's something Microsoft is calling an Autopilot: an always-on, autonomous agent with its own identity that acts on your behalf without waiting for a prompt.
Most "AI at work" today still waits for you to ask. Autopilots don't wait. They stay active in the background, understand how work gets done across your apps and systems, and take action continuously under the permissions and policies you set.
What Microsoft Scout actually does:
- •**Proactive scheduling** — monitors traffic and your calendar to recommend when to leave for appointments, school pickups, dinner dates
- •**Meeting coordination** — schedules and coordinates across time zones, flags important meetings, generates prep materials
- •**Deliverable tracking** — identifies upcoming deliverables and automatically blocks calendar time to keep you on track
- •**Risk surfacing** — spots stalled decisions and potential blockers before they become problems
- •**Background awareness** — reads Teams threads, transcripts, and email to surface what matters to you without you asking
- •**Out-of-office management** — figures out dates, checks calendar conflicts, and creates blocks automatically
Where it lives:
Scout operates across cloud, desktop, and web. It connects to Teams, Outlook, OneDrive, SharePoint, and your local device resources. You interact with it in Teams, and extend its reach through a desktop app to your browser, local resources, and MCP servers.
The security model:
Every Scout agent operates under its own governed Entra identity — not a shared, anonymous service account. Admins set policy rules. Sensitive actions require human sign-off. Microsoft Purview sensitivity labels and DLP policies are enforced in real time. Credentials are scoped to the task, redacted from logs, and managed with first-party rigor. Microsoft Scout doesn't bypass your controls; it operates within them.
Built on OpenClaw:
In a significant move, Microsoft is contributing policy conformance directly upstream to OpenClaw. Organizations running OpenClaw will be able to validate whether their environment meets security and compliance requirements with a verifiable, audit-ready answer. Scout is powered by open-source OpenClaw technology with Work IQ as its context engine.
Availability:
Microsoft Scout is available through the Frontier program as an experimental release. Access requires Frontier enrollment, Intune policy configuration, an opt-in attestation, and a GitHub Copilot license. More than 3,000 Microsoft employees are already using the desktop experience internally. A broader cloud preview will roll out in the coming months.
Omar Shahine, CVP of Microsoft Scout, put it directly: *"This is a personal assistant, it's the first real personal assistant we've offered customers... you're going to get a phone call from this assistant, it's a very different type of AI than chat."*
Majorana 2: Quantum Computing Accelerated
Microsoft's quantum team announced Majorana 2, the next generation of its topological quantum chip. The qubits are now 1,000x more reliable than the previous generation, with a mean lifetime of 20 seconds — and instances lasting up to one minute. Other quantum approaches measure qubit lifetime in microseconds.
The team used advances in agentic AI (via Microsoft Discovery) to rapidly improve topological qubits. The combination of reliability, speed, and small size sets the stage for a commercially relevant quantum computer by 2029.
Microsoft Discovery Goes GA
[Microsoft Discovery](https://azure.microsoft.com/en-us/solutions/discovery), the agentic AI platform for R&D, is now generally available. The Discovery Engine uses specialized agents to mimic the scientific method across large knowledge bases — generating hypotheses and validating theories in a continuous loop. An early preview of the Microsoft Discovery app also launched, offering a local version that runs on researchers' own machines with just a GitHub Copilot account.
MAI: Microsoft's In-House Model Family
The Microsoft AI Superintelligence Team unveiled its newest generation of in-house models:
- •**MAI-Thinking-1** — a 35 billion active parameter reasoning model built from the ground up on clean data, without distillation from third-party models. Designed for complex multi-step instructions, long-context reasoning, and code generation. Now open to select early partners.
- •**MAI-Image-2.5** — Microsoft's first model serving both text-to-image and image-to-image workloads
- •**MAI-Transcribe-1.5** — state-of-the-art accuracy with entity biasing; streaming coming soon
- •**MAI-Voice-2** — now available in 10+ additional languages with new voice options
- •**MAI-Code-1-Flash** — purpose-built for GitHub Copilot and VS Code
Image, transcription, and voice models are generally available now on Microsoft Foundry and MAI Playground.
Frontier Tuning: Your Data, Your Model
Frontier Tuning lets organizations train domain-specific models on their own data, workflows, and style — without adapting their work to a generalized foundation model. It runs on a single platform inside your own tenant, ships with enterprise reinforcement learning environments, and you own both the model and the learning loop. Now open to select early partners.
GitHub Copilot App
The GitHub Copilot app is a native desktop experience for agentic development. It uses git worktrees so each session has isolated branches, files, and task state. You can start from existing GitHub Issues and Pull Requests, pause and resume sessions, and orchestrate multiple agent sessions in parallel. Available in technical preview for Copilot Pro and higher tiers.
Rayfin: Backend in One Command
Rayfin is an open-source SDK and CLI that generates typed, governed backends (database, auth, storage, access policies) from a natural language description. One CLI command deploys to Microsoft Fabric as a managed service inheriting your tenant's security and governance. Microsoft partnered with Replit so you can build in Replit and deploy with Rayfin, keeping everything inside your own Fabric tenant.
Microsoft Foundry: The Production Layer
Foundry's latest releases round out what production agents have been missing:
- •**Hosted Agents** — per-session sandboxing, sub-100ms cold starts, zero idle cost. GA by end of June.
- •**Microsoft Agent Framework 1.0** — GA today. Drop GitHub Copilot SDK or Claude Agent SDK agents into MAF workflows.
- •**Foundry Toolboxes** — unified access to web/file search, MCP, OpenAPI, A2A protocol
- •**Fireworks AI on Foundry** — GA, single Azure endpoint for open-source models
- •**Agent Control Specification** — define and enforce what agents can do in production
- •**Adaptive Evaluations** — convert policies into automated agent behavior tests
- •**Agent Optimizer** — turns evaluation signals into ranked candidate improvements
What This All Means
Microsoft Build 2026 wasn't a collection of feature announcements. It was a statement of strategic direction: the future of computing is agentic, and Microsoft intends to own the full stack.
From the silicon (RTX Spark) to the OS (Windows Agent Framework, Project Solara) to the intelligence layer (Microsoft IQ) to the agents themselves (Scout, Autopilots) to the tooling (Foundry, GitHub Copilot app) to the governance (Agent 365, Entra identities, Purview enforcement) — Microsoft is building a vertically integrated agent platform.
The OpenClaw connection is particularly telling. Microsoft isn't just consuming open-source agent technology; it's contributing upstream, building enterprise-grade security on top of it, and shipping it as a first-party product in Microsoft Scout. After Satya Nadella's earlier skepticism about OpenClaw, this is a meaningful pivot — and a bet that the open agent ecosystem will be as strategically important as the open web was two decades ago.
For developers: the Windows Agent Framework is open source and ready to build on. For IT leaders: Agent 365, Microsoft Scout's governance model, and the consumption-based Copilot Credits pricing give you control and visibility. For end users: the first always-on personal assistant that actually understands your work is here — if you're in the Frontier program, you can try it today.
The agent era isn't coming. Microsoft just shipped it.
*Sources: Microsoft Build 2026 Live Blog, The Verge, Microsoft 365 Blog, A Guide to Cloud*