Microsoft Build Live
That’s a wrap! As you can see, there’s plenty to be excited about as Microsoft builds the new toolchains and software for this era of development. We had a great time sharing all the exciting developments announced today. Huge new products, major updates and more intuitive experiences are at the core of these Build announcements, all designed to help developers build, create and innovate. Remember, you can visit the Build 2026 homepage for more info on sessions, demos and to hear additional speakers. Also, make sure to check out the Official Microsoft Blog and learn more about all the news announced during Build with the GitHub Copilot CLI . Thank you for tuning into this keynote and our live blog. Introducing Majorana 2, Microsoft’s next-gen quantum chip Last year, Microsoft Quantum released the Majorana 1 chip, which relied on a new state of matter that previously existed only in theory. At the time, the team said the long-term vision for quantum was to take a radical approach in addressing the fundamental barriers that have limited a scalable quantum machine: size, speed and reliability. That’s why the release of Majorana 2 is so exciting — it’s actually a huge step towards that goal of a million qubits on a single chip that can fit in the palm of the hand. The quantum team took advantage of recent advances in agentic AI to rapidly improve its topological qubits, and as a result, they’ve overcome barriers that have limited the application of quantum computing to real-life scenarios in the past. The qubits powering Majorana 2 are now 1,000x more reliable than the previous generation and are capable of maintaining their quantum state much longer. Other common quantum approaches measure a qubit “lifetime” in microseconds, but get this: Majorana 2 offers a mean qubit lifetime of 20 seconds, with instances lasting as long as one minute! This combination of reliability, fast speed and small qubit size sets the stage for the quantum team to achieve a scalable quantum computer that is commercially relevant by 2029. The promise of this type of machine is enormous (think solving intractable problems in health care, energy, agriculture and other industries) and yeah, it still sounds a little like science fiction. But with Majorana 2 and other advancements by the quantum team, it’s getting closer and closer to reality. – Christina Microsoft Discovery now generally available Microsoft Discovery , an agentic AI platform focused on a new era of research and development, is now generally available. At its core is the Discovery Engine, which uses specialized agents to mimic the scientific method across large amounts of knowledge to generate hypotheses and validate theories in a continuous loop. Plus, there are built-in controls for enterprise security, transparency and governance to help accelerate scientific and engineering research. This is especially useful for customers in highly demanding and regulated industries like food and energy, because they now have the ability to utilize agentic discovery for their R&D. The goal is to accelerate the pace of scientific discovery across industries. Agentic AI can be used here to optimize experiments, manage workflows, propose new solutions or pinpoint previously unnoticed flaws in the research process. Researchers can deploy autonomous agent teams, guided by human expertise. And it gets better! There is a new early preview of the Microsoft Discovery app, which is designed to expand access to this cutting-edge agentic AI for researchers, from students and individual scientists to academic labs and public research institutions. The app offers a local version of Microsoft Discovery’s core capabilities that can download and run from their own machines. You just need a GitHub Copilot account to get started. – Christina The next-gen in-house Microsoft AI models This one is exciting! The Microsoft AI Superintelligence Team recently finished the newest generation of its in-house models, which now support reasoning, code generation and text-to-image and image-to-image workloads for developers. MAI-Thinking-1 is Microsoft AI’s first reasoning model. It is a mid-sized, 35 billion active parameter model built for high efficiency and performance, but importantly, at a low-token cost (just say “no” to tokenmaxxing!). MAI-Thinking-1 was designed to be good at complex multi-step instructions, long-context reasoning and code generation. Notably, it was built from the ground up on clean data, without distillation from third-party frontier models. It’s designed to be fast, cheap, reliable and highly performant, and it is now open to select early partners. But that isn’t the only new model! MAI-Image-2.5 is Microsoft’s first model to serve both text-to-image and image-to-image workloads. This is especially useful in developer workflows, when you want some assistance taking a concept into reality or you want to enhance some existing image work. There are other new members of the MAI family too: MAI ‐ Transcribe ‐ 1.5 combines state-of-the-art accuracy and entity biasing, with streaming coming soon. MAI-Voice-2 is now available in more than 10 additional languages with new voice options. And MAI-Code-1-Flash is purpose-built for GitHub Copilot and VS Code to deliver high performance and lower cost. Image, transcription and voice models are generally available now on Microsoft Foundry and MAI Playground. – Christina Frontier Tuning gives companies domain-specific, high-efficiency AI We’ve heard from many developers who want agents that are tuned to the work that matters most to them, built using their data, workflow and style. Frontier Tuning is designed to help you stop adapting your work to the model and allow you to train domain-specific models, which can help you further optimize reinforcement learning and increase efficiency. Frontier Tuning is about getting more out of the models. Foundation models, while powerful, are highly generalized — they don’t know your APIs, your terminology or your internal processes. Tune models custom fit to your domain, workflows, policies and house style, which can yield higher-quality results faster and at lower cost. Frontier Tuning runs on a single platform running inside your own tenant. It also ships with enterprise reinforcement learning (RL) environments. Your AI learns directly from real workflows — think task completions, decision reviews, validated outcomes. You own the model and the learning loop. We are nowopening Frontier Tuning to select early partners. – Nick Microsoft Scout, the new always-on personal agent for work built on OpenClaw Most of what we call “AI at work” today still waits for a prompt. With Autopilot agents, that work is becoming more continuous. Autopilots are always-on agents that work autonomously with their own identity, acting on your behalf. They stay active in the background, understand how work gets done across your apps and systems and take action without needing to be prompted each time. Microsoft Scout is Microsoft’s first Autopilot agent. It is built with enterprise-grade security and powered by open-source OpenClaw technology, with Work IQ as its context engine. It lives where work already happens: Teams and Outlook for conversation, OneDrive and SharePoint for files, plus device-local actions on your machine. Microsoft Scout’s trust layer is built in, not bolted on. Every agent operates under its own governed Entra identity, not a shared, anonymous service account, so the work it does is always attributable to a known actor your directory already understands. Admins will be able to set policy rules, and you can see what the agent is doing in the background as work progresses. I’ve been testing an early local version of Microsoft Scout for several weeks before today. One of my favorite things: I set it up to monitor a GitHub discussion every morning, resolve the right feature owners across Microsoft 365 apps and open Teams chats to track status — without me lifting a finger. And after weeks of tokenmaxxing to get all these announcements ready, I’m going to need a break — so I asked Microsoft Scout to set my out-of-office for the week after Build. It figured out the dates, checked my calendar for conflicts and created the block. Work IQ is what makes that useful instead of noisy — it learns the people, files and patterns specific to how you work. Microsoft Scout is available to Frontier organizations through an early experimental release, giving customers a chance to explore how Autopilots can fit into their own workflows. Access requires Frontier enrollment, Intune policy configuration and an opt-in attestation. Users with a GitHub Copilot account/license can then download and install the experience. – Nick A new look for Microsoft Copilot Next up, Satya is discussing the new design for Copilot that was recently announced. Check this out for more details. And now ... The Chainsmokers! Here are the special guests for today’s event, the Grammy-winning duo The Chainsmokers. If you’re here at Build, catch them live at a special concert at 6 p.m. PT here at the Fort Mason Center in San Francisco. The latest on MDASH Right now, Satya is talking about MDASH , the new multi-model agentic security system that was announced last week. Learn more about it here . Rayfin delivers enterprise-grade app backends Coding agents can generate an app in seconds. Getting it to production is still the slow part — wiring up the backend plumbing before a single user touches it. Rayfin , now in preview, collapses that work. Rayfin is an open-source SDK and CLI. You describe the app you want — in code or in natural language to a coding agent — and Rayfin will generate a typed, governed backend: database, auth, storage and access policies. One CLI command ships it to Microsoft Fabric, where it runs as a managed service and inherits the security, governance and compliance controls your tenant already enforces. App data lands in OneLake by default — no copy, no ETL. We’ve partnered with Replit, a leading AI coding agent platform where devs turn natural language into running apps. Build in Replit, deploy with Rayfin and the app, data and services stay in your own Fabric tenant — under the identity, network and governance controls you already trust. – Nick The GitHub Copilot app for agentic development Agents let us write code faster than ever before, but that’s only one part of building software. The GitHub Copilot app —which is now available in preview — is a native desktop app that brings the best of agentic development to the place you’re already spending a ton of your time: GitHub. I’ve been using this app for a while and I love it, because instead of just starting with a blank prompt or a bunch of disconnected tools, you can start from existing GitHub Issues, Pull Requests and even other sessions. And those details and context stay with you and stay connected to your session. Under the hood, the GitHub Copilot app uses git worktrees, so each session has its own space with its own branches, files, conversations and task state. This way, the work you’re doing stays separated, even when you’re working on multiple things at a time. And you can pause and resume sessions, so you can come back to where you left off. The app also makes it easy to orchestrate multiple agent sessions in parallel and keeps changes moving through validation, review, CI and merge. Copilot handles the execution. Developers stay in control. It’s the agent-native way to use GitHub. – Christina Microsoft Foundry updates: agents you deploy, improve and trust in production The hard part of building agents isn’t the prototype — it’s everything after. Isolation. Identity. Tools that actually work. A path from production back to a better version. Microsoft Foundry’s latest releases round out the four layers production agents have been missing. Hosted Agents in Foundry Agent Service will be generally available in coming weeks, providing per-session sandboxing for untrusted code, sub-100 ms cold starts and zero idle cost in a framework-agnostic runtime. Build. Microsoft Agent Framework (MAF) v1.0 (generally available) ships an agent harness as a first-class concept — skills, context, memory and middleware are production-ready. Drop a GitHub Copilot SDK or Claude Agent SDK agent into a MAF workflow as a named participant; the orchestrator stays deterministic. New toolboxes in Foundry (preview) will unify access to capabilities like web and file search, MCP, OpenAPI spec and A2A protocol. Fireworks AI on Foundry (generally available) offers fast access to open-source models through a single Azure endpoint with Managed Compute (private preview) adding dedicated GPU capacity. Ground. Foundry IQ (generally available) knowledge bases unify Work IQ, Fabric IQ, File Search, Azure SQL and MCP behind one SLA-backed retrieval endpoint. Microsoft World Grounding extends that reach to live external information alongside internal data. Procedural memory (preview) will let agents learn the “how” across multiple runs, not just the “what.” Operate. Tracing and evaluation for hosted agents are generally available — one OpenTelemetry pipeline, evals linked back to the trace that produced them. Agent optimizer (preview coming soon) in Foundry Agent Service will turn those signals into ranked candidate improvements across prompt, tools, skills and context, with diffs, audit and one-click rollback. Adaptive Evaluations (preview) will convert policies into automated tests for agent behavior — usable standalone or built into Foundry. Agent Control Specification (preview) will let teams define and enforce what agents can do in production, across Foundry, Microsoft Agent Framework and LangChain. Reach . One-click publishing to Microsoft Teams and Microsoft 365 Copilot will be generally available next month. Identity and tenant policy will flow through automatically. – Nick Windows 365 for Agents brings Cloud PCs to agent workloads Every dev these days is using agents and every engineering manager wants their devs to use MORE agents. That makes it critical for organizations to have a way to create secure, scalable agent execution environments. This is where Windows 365 for Agents comes into play. Agent makers can use it as part of Agent 365 tools, now generally available, or through a preview available in Microsoft Copilot Studio. It enables enterprise automation by giving AI agents secured and managed Cloud PCs that operate inside the environments where business happens. This way, agents can interact directly within applications and browsers, execute multi-step workflows and use the same systems that enterprises rely on every day, whether modern, legacy, API-based or UI-driven. And because every Cloud PC is Entra ID-joined, Intune-managed and policy-enforced, it gives IT a consistent way to scale agents without compromising compliance or security. And agents of all stripes are supported by Windows 365 for Agents,
Source: Microsoft Source