On June 2, local time in the United States, the Microsoft Build 2026 Developer Conference kicked off at the Masonic Temple in San Francisco. The theme of this conference focused on the practical application of cutting-edge AI technology. Microsoft unveiled a series of products and updates covering self-developed AI models, intelligent agent applications, operating system security, developer tools, cloud services, and new hardware platforms.
At the 2025 developer conference, Microsoft established the direction of the "AI Agent Era," releasing Copilot Studio for multi-agent orchestration, Windows AI Foundry, and announcing full support for the Model Context Protocol. GitHub Copilot introduced the programming intelligent agent Coding Agent.
In Microsoft's narrative, 2025 addressed "what standards and frameworks to use in the Agent Era," while 2026 focuses on "how to truly operationalize using their own models and products." The model layer has been reinforced with key self-developed models, and the product layer has transitioned the intelligent agents from demonstrations to a full-stack implementation across systems, hardware, and the cloud.
This conference's core releases can be divided into six main sections: MAI Self-developed Model Family, the intelligent agent ecosystem represented by Scout and GitHub Copilot applications, Windows system-level AI security sandbox MXC, the developer-oriented Surface RTX Spark Dev Box and system optimizations, Project Solara new intelligent agent device platform, and developer tools and governance frameworks including Microsoft IQ, Rayfin, ASSERT, and ACS.
01 Seven Models Trained from Scratch, Rejecting Distillation
The entire keynote speech unfolded slowly along the vision statement of Microsoft CEO Satya Nadella. After presenting the "Agent-First" strategic framework, executives from various business lines took the stage one by one, launching specific products to implement this framework.
At the conference, Suleiman announced the release of seven brand-new models developed internally by Microsoft AI, all unified under the MAI family.
He described MAI's mission as building a "climbing machine" that achieves continuous self-improvement through ongoing investments in computation, better data, and more precise evaluation—a loop of self-improvement, keeping users constantly at the forefront of technology.
In terms of training scale, Suleiman pointed out that the compute used to train cutting-edge models has increased by a trillion-fold and is expected to grow another thousandfold in the next three years. All Microsoft MAI models undergo "hill climbing from scratch, zero distillation" and do not rely on third-party model outputs for training.
The specific models are as follows:
Flagship Inference Model MAI-Thinking-1, which is a medium-sized model. Microsoft states that it performs on par with the best models on the market in key software engineering tests. In a blind comparison, human judges showed a preference for it comparable to Sonnet 4.6. This model is trained from scratch with clean data and does not use third-party model distillation.
Coding Model MAI-Code-1-Flash, an inference-efficient agentic coding model with 5 billion parameters, tailored for GitHub Copilot, VS Code, and the Microsoft stack with deep integration. Microsoft claims it can rival Haiku at a lower cost.
Visual-Natural Language Model MAI-Image-2.5 and its ultra-efficient Flash variant, supporting visual-natural language and image editing. Microsoft states it surpasses Google Nano Banana Pro in Arena scores.
Transcription Model MAI-Transcribe-1.5, with state-of-the-art accuracy. It is reportedly five times faster than competitor models and includes built-in support for recognizing domain-specific terms in 43 languages.
Voice Generation Model MAI-Voice-2, providing high-quality, natural-sounding voice generation supporting 15 languages, capable of adapting voices based on short samples, and equipped with anti-abuse measures. Its Flash variant is about to be released to achieve the same functionality at a lower cost.
All models share the same data specifications, infrastructure, and evaluation framework. In addition to distribution on Azure Foundry and optimization for Microsoft first-party products, these models will also be available to developers on Open Router, Fireworks, and Baseten. Developers will now be able to adjust model weights themselves for the first time.
At the event, Nadella introduced Microsoft Frontier Tuning, a method for enterprises to customize models using their own work data. The rationale is that the most valuable data is not general language corpora but the real trajectories, steps, and decisions of intelligent agents performing tasks within the enterprise.
This framework integrates the MAI model into actual business processes, allowing the model to learn on the fly in a real-world environment. Suleiman said, "You are building your own model: in your environment, using your data for training, under your control. Your institutional knowledge becomes part of the model and belongs only to you."
In terms of performance, the MAI model fine-tuned by Microsoft for Excel is on par with GPT-5.4, while achieving a 10x efficiency improvement. After adopting Frontier Tuning, McKinsey achieved the highest win rate among all test models, with costs reduced by approximately 10x.
In the healthcare field, Microsoft announced a collaboration with Mayo Clinic to jointly develop a cutting-edge AI model for healthcare. This model will combine Mayo Clinic's clinical expertise, de-identified clinical data, and longitudinal insights with Microsoft's core AI capabilities.
Microsoft also revealed that the MAI model is co-designed with its in-house Maia 200 chip, achieving a 1.4x efficiency improvement through software-hardware co-optimization.
02 Comprehensive Deployment of Intelligent Agents Ecosystem
Microsoft announced at the conference a grand transformation towards "Agent-first," aiming to automate how knowledge workers use software and embed AI assistants into everyday office interactions.
Scout is the core intelligent agent product released this time. This AI Agent, known as "always-on," is built on the OpenClaw framework and can interact in Microsoft Teams like a human colleague.
Scout can browse through a user's work messages, calendar, and email inbox, automatically complete tasks, reschedule conflicting meetings, and draft professional-sounding replies. Users can send direct commands to it in Teams or give it a personalized name.
Microsoft's newly appointed Corporate Vice President Omar Shahine explained the design philosophy behind Scout: "Your company essentially hires your assistant. The whole point of having a personal assistant is that when you're not working, they are."
Scout is provided through the Microsoft Frontier program and requires a GitHub Copilot subscription. Microsoft is currently testing a Scout desktop application that will be rolled out to subscription users with access to select "frontier" features. Within Microsoft, Shahin said the sales department is the largest and fastest-growing group using the tool.
The GitHub Copilot desktop application is another key release. Mario Rodriguez, Chief Product Officer at GitHub, introduced it as a "GitHub-native, Agent-native desktop experience."
Through a unified "My Work" view, developers can see dynamic work across interconnected repositories, including activity sessions, issues, pull requests, and background automation.
Each session runs in its own Git worktree, and parallel Agents do not interfere with each other. The application features an Agent Merge capability that guides pull requests through reviews, checks, and merges. A Canvas interface is used for bi-directional interaction between humans and machines, allowing developers to inspect, guide, and validate the work performed by the Agent on their behalf.
The GitHub Copilot application is available for technical preview on Windows 11, Windows 11 on Arm, Mac, and Linux, requiring a GitHub Copilot subscription and will be made available to Copilot Free users in the future. The application supports both cloud and local sandbox environments, with policy support included for both.
In terms of AI governance, Microsoft has released the Agent Control Specification (ACS), a new open standard aimed at providing developers with a more consistent and fine-grained way to control AI Agent behavior. ACS enables development, compliance, and security teams to define policy files for Agents, specifying what Agents can and cannot do, when human approval is required, and what evidence should be recorded for review.
ACS is released as an SDK and includes plugins for LangChain, OpenAI Agents SDK, Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI, and MCP tools, among others. Since policies can be written as a single file, they can be bundled with the Agent, following the Agent across different frameworks and environments.
ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing) is another testing tool. This is an open-source framework that converts a high-level natural language description of a target, policy, or expected behavior into structured scoring tests using AI.
ASSERT takes a concise language description of the expected behavior of an AI model, generates sets of acceptable and unacceptable behaviors, problem scenarios, and test cases, runs tests against the target system, and scores them. It also records the path taken by the AI system, including intermediate operations and tool invocations, for developers to inspect failure points.
03 The more autonomous the Agent, the more dangerous, Microsoft draws a line at the system level with MXC
As AI Agents become more powerful and autonomous, Microsoft has identified a key issue: the more autonomous the Agent, the more useful it is, the more dangerous it is to allow it to run unrestrained on an enterprise network. Microsoft's official blog describes this as a "multi-layer systemic issue," where every interaction between the Agent and humans, tools, applications, models, and other Agents "exposes new attack surfaces and introduces different failure modes."
To address this issue, Microsoft has introduced Microsoft Execution Containers (MXC), which is a policy-driven execution layer built into the Windows operating system itself.
Pavan Davuluri, Corporate Vice President of Microsoft Windows and Devices Execution, emphasized that this is crucial to making AI Agents commercially viable, focusing on security, containment, isolation, and user control, making the Agents safe enough for deployment to both regular consumers and enterprises.
MXC is essentially an SDK and policy model embedded in Windows and Windows Subsystem for Linux, providing what Microsoft calls a "composable sandbox spectrum." This spectrum ranges from lightweight process isolation (already adopted by GitHub Copilot's command-line interface) to micro virtual machines, Linux containers, and full cloud instances running on Windows 365.
The system separates Agent execution from the user's desktop, clipboard, UI, and input devices. Each Agent is tied to an identity, either a local ID or a cloud-provisioned identity supported by Microsoft Entra, ensuring that every action of the Agent can be attributed, audited, and governed.
MXC is now available in an early preview. Agent 365, integrated with the Microsoft Enterprise Security Stack, will launch a preview in July 2026, layering Entra identity services, Intune device management, Defender threat protection, and Purview data governance capabilities on top of MXC, enabling IT departments to centrally manage Agent isolation.
On the partnership front, OpenAI, NVIDIA, Manus, Nous Research (manufacturer of the Hermes Agent), and the OpenClaw open-source project have announced their plans to build on MXC.
Notably, the collaboration with OpenClaw began when its creator, Peter Steinberger, proactively reached out to Microsoft expressing interest in collaboration, which eventually evolved into a comprehensive platform-level partnership.
04 Three Updates Making Edge AI "Offline-First"
Microsoft Edge browser has also received a local AI capability upgrade. Microsoft stated that since the introduction of Phi-4-mini in Build 2025, the team has expanded edge AI capabilities based on web developer feedback.
The first is Aion-1.0-Instruct, a smaller, faster, and more efficient local small language model than Phi-4-mini.
It can run on PCs with weaker GPU and CPU capabilities, now available in a developer preview and set to land on Hugging Face in July.
The second is the Language Detection and Translation API, available with Edge version 148. These two APIs are powered by Edge's built-in edge AI models, used in JavaScript to allow websites and browser extensions to detect text language and translate between language pairs.
Microsoft called it "providing fast, high-quality translations, supporting over 145 languages, and optimized for translation workloads online," with this service being free.
The third item is to implement speech recognition through the Web Speech API, offered in an experimental form in the Edge Canary and Dev channels.
This API helps developers integrate speech or audio input into websites and browser extensions, running locally on the device, and can also leverage cloud-based speech-to-text and text-to-speech services as a fallback.
05 Developer Tools and Cloud Service Iteration
On the data intelligence front, Microsoft has released Microsoft IQ, consolidating the previously separate four contextual sources into a shared foundation called Agent.
Microsoft Fabric's Chief Technology Officer, Amir Netz, made an analogy: the cascading green code in "The Matrix" is not just decoration but the foundation of that world. He said, "What we're doing in the data world is creating a data-based reality for Agent."
The four contextual sources for Microsoft IQ are: Work IQ, capturing organizational day-to-day operations through email, documents, meetings, and schedules; Foundry IQ, managing institutional knowledge by planning and indexing knowledge bases; Fabric IQ, modeling real-time operational status of the business through data, defining entities, relationships, and business rules anchored by real-time signals based on Fabric, a feature expected to be formally released in the coming months; Web IQ, adding real-time global context from the web.
With this contextual framework, Agent is no longer just a tool that executes commands but a virtual employee that understands the company's operations.
Having a shared "foundation" alone is not enough. As Agent starts generating applications, each application needs a backend. If left unattended, these applications will form new data silos outside the contextual layer. To address this, Microsoft has released Rayfin, an open-source SDK and CLI that directly deploys applications built by Agent to the Fabric platform as governed production backends. Application data defaults to enter a unified OneLake data lake, feeding back to Microsoft IQ, instead of accumulating externally.
Microsoft positions it as a competitor to Supabase and Neon, with the core difference being governance: all applications go through the same set of data and compliance channels. Netz described this as a two-way process where Agent retrieves information from the enterprise's data rules when building an app, and the data generated when the app runs updates these rules, enabling the next Agent to leverage the latest information.
Microsoft also introduced the WSL Container feature, allowing developers to create and manage Linux containers directly on Windows. Microsoft has equipped it with a command-line interface and API, enabling the running of Linux containers within native Windows applications. This feature will be available for public preview in the coming months.
To save developers time on environment configuration, Microsoft has also released Windows Developer Configurations. This tool allows for the quick setup of a new machine with developer-optimized configurations. It automatically installs WSL, PowerShell 7, and Visual Studio Code, while enabling Git version control in File Explorer and displaying hidden files.
06 Two New Hardware Releases Bringing AI Back to the Edge
This Build event was not just a showcase of software such as models, agents, and development tools; hardware also had its time in the spotlight. As AI computing demands more power and Agentic workflows require continuous processing, Microsoft turned its focus to the devices developers have at hand. Instead of relying on expensive cloud GPUs every time, why not let these tasks run directly on local machines?
Microsoft's Corporate Vice President of Surface Devices, Andrew Hill, announced two new devices:
The Surface RTX Spark Dev Box is a compact developer PC equipped with the NVIDIA RTX Spark superchip. It combines an NVIDIA Blackwell RTX GPU and an NVIDIA Grace CPU, delivering up to 1 Petaflop of AI computing power and is equipped with 128 GB of unified memory.
This device features an aluminum chassis that doubles as a heatsink, designed for long-running training tasks, large model inference, and complex Agentic processes.
The device comes preloaded with Windows 11 Pro and is preconfigured for developers at the image level: dark theme, taskbar optimized for development, widgets removed, 'Do Not Disturb' mode enabled, developer mode activated, PowerShell 7 as the default shell. WSL 2 is configured with GPU pass-through and CUDA support, and VS Code, GitHub Copilot, Git, Python, and Node.js are all installed.
Regarding security, the Surface RTX Spark Dev Box is built on chip-to-cloud security that adheres to Microsoft's Zero Trust principles. This includes the Secured-core PC architecture, BitLocker encryption, and Microsoft Defender protection. It can be integrated with Entra ID and Intune for large-scale management and governance.
Hill explained, "The way developers build software is undergoing a fundamental change. The capabilities and complexity of AI models are constantly growing, Agentic workflows require continuous computing power, and even tasks that do not require cutting-edge models may incur cloud costs with each iteration."
Another Surface device, the Surface Laptop Ultra, designed for developers, creators, and technical professionals, has already been released earlier, both representing the next step for Surface: creating dedicated devices for building the future. The Surface RTX Spark Dev Box will be launched later this year in the U.S., exclusively available on Microsoft.com.
07 A New Platform for Running AI Agents on Devices Instead of Apps
Stevie Batiste, Head of Microsoft's Applied Sciences division, introduced an internal project called Project Solara.
This is a new platform from chip to cloud, based on Android instead of Windows, designed to run AI Agents on devices instead of apps. Batiste explained its rationale: "The boundaries are collapsing. You don't necessarily need a traditional app model. You don't need traditional ways to develop experiences."
The first two concept devices were showcased at the Build conference:
A desktop hub device, placed next to a PC, responds to voice commands, logs in users through facial recognition, and presents the day's most urgent items. When connected to a display, it transforms into a full Windows machine running in the cloud.
A wearable badge device, reimagining the standard employee ID card. A single fingerprint press wakes up the Agent, a light touch records and transcribes conversations, and a built-in camera allows the Agent to take action based on what the user sees.
In a healthcare demo, this badge ran an Agent designed for medical staff, able to scan patient QR codes, record and transcribe medical visits, capture vital signs, and prescribe medication. In another scenario, the built-in camera scanned a brainstorming board with office renovation ideas and suggested adding plants.
Batiste stated that Microsoft will not produce these devices themselves but envisions hardware manufacturers and other industry partners transforming these reference designs into their own products, each tailored to specific industries, companies, or scenarios.
08 Quantum Chip Upgrade, Reliability Improved Thousandfold
Microsoft also unveiled the next-generation topological quantum chip, Majorana 2.
Compared to its predecessor Majorana 1, the key change this time is the superconductor material being switched from aluminum to lead. This adjustment has increased the quantum bit reliability by 1000 times, with an average quantum bit lifetime reaching 20 seconds, and in some instances extending to a minute.
The quantum bit lifetime of other technical paths is usually only on the order of microseconds. Based on this advancement, Microsoft has halved the expected time to achieve a scalable quantum computer, with the current estimate set for before 2029.
The development of this chip utilized the Agentic AI capability of the Microsoft Discovery platform throughout. The AI agent performed tasks such as manufacturing management, automated measurement of quantum states, and interdisciplinary data analysis, compressing the measurement cycle from several weeks to several orders of magnitude less and identifying correlations imperceptible to humans from nearly two decades of accumulated data.
Says Microsoft Technical Fellow Chetan Nayak: "Agentic AI permeates almost everything we do." However, he emphasizes that AI only provides guidance, "scientists are always in the loop."
The Microsoft Discovery platform was also officially launched at this conference, which is an organization-wide platform for cutting-edge research and allows researchers to deploy human-guided autonomous Agent teams for hypothesis generation, experiment optimization, and theory validation. Microsoft has simultaneously released an early preview of the Microsoft Discovery app, which individuals can download for free and run locally using their GitHub Copilot accounts.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
