Microsoft Build Conference announces the entry into the era of AI agents, Microsoft 365 Copilot and GitHub coding upgrades, Musk's xAI model is included in Microsoft Cloud

Source: Wall Street News

At its annual Build developer conference, which began on Monday, May 19 (Eastern Time), Microsoft demonstrated that it is transforming Windows into a core platform for AI agents at an unprecedented pace. By natively supporting the Model Context Protocol (MCP) and launching Windows AI Foundry, Microsoft is paving the way for the automated future of AI agents.  

During Monday’s two-hour keynote speech, Microsoft comprehensively showcased how to upgrade "AI assistants" to "AI development partners" and drive the full entry of AI applications, development, and enterprise operations into the "agentic" era. This Build conference represents Microsoft’s full embrace of AI-native development, intelligent agents, and the reconstruction of the Windows ecosystem. The core announcements include:  

### GitHub Copilot Upgraded to "Autonomous Programming Agent"  

Microsoft 365 Copilot Introduces Copilot Tuning Capabilities for Multi-Agent Collaboration  

Azure AI Foundry Adds xAI Models  

Windows 11 Becomes an AI-Native Operating System  

Edge Browser Transforms into an AI Web App Platform  

Launch of AI Developer Platforms like Copilot Studio and Windows AI Studio  

Building the Multi-Agent Agentic Web and NLWeb Project  

#### GitHub Copilot Programming Agent and GitHub Models Upgrade  

GitHub Copilot is evolving from an assistant built into code editors to an AI intelligence partner with agentic capabilities, and for the first time, it integrates asynchronous code agent functions within the GitHub platform. GitHub Models also introduces new prompt management, lightweight evaluation mechanisms, and enterprise-level control functions, allowing teams to flexibly test these models on the platform without leaving GitHub.  

Additionally, Microsoft has open-sourced GitHub Copilot Chat in VS Code, incorporating the AI functions of the GitHub Copilot extension into the open-source repository of this world’s most popular development tool. As a technology platform with over 150 million developers, these moves further strengthen Microsoft’s long-term commitment to open, collaborative, and AI-driven software development models.  

#### Microsoft 365 Copilot Tuning and Multi-Agent Collaboration Features  

The newly launched features enable enterprise customers to train models and build agents using their own corporate data, workflows, and procedures through the new Copilot Tuning function in a simple, low-code manner.  

These agents can run securely within Microsoft 365 services to perform highly accurate, domain-specific tasks. For example, a law firm could create an agent to automatically generate legal documents that match its professional field and writing style.  

Furthermore, the new multi-agent orchestration function introduced in Copilot Studio allows multiple agents to collaborate, combining their respective expertise to handle broader and more complex tasks.  

#### Grok 3 and Grok 3 mini Added to Microsoft Cloud Azure AI Foundry Model List  

Microsoft officially launched Azure AI Foundry models and new model evaluation tools. Azure AI Foundry is a unified platform designed for developers to design, customize, and manage AI applications and AI agents.  

Microsoft announced that it will add AI models from Elon Musk’s AI startup xAI to its cloud computing platform. The Azure AI Foundry model list now includes xAI’s Grok 3 and Grok 3 mini, which are directly hosted and billed by Microsoft and provided to Microsoft’s own product teams and customers through the Azure AI Foundry service.  

"These models will have all the service level agreements (SLAs) that Azure customers expect from any Microsoft product," the company said.  

Microsoft noted that users of its Azure cloud service can access more than 1,900 AI models, including those from partners like OpenAI, Meta, and DeepSeek. The addition of xAI models expands users’ options. However, media outlets have pointed out that some industry "stars" such as Google and Anthropic models are conspicuously absent from Microsoft’s cloud platform. Microsoft is competing with rivals like Amazon and Google in selling and leasing computing power to gain a foothold in the AI application building and deployment platform space.  

Microsoft also introduced new tools, including Model Leaderboard, which evaluates the best-performing AI models for different tasks and categories, and Model Router, which automatically selects the most suitable model based on real-time query or task requirements.  

### Windows AI Foundry: Unified Local AI Development Platform  

At the Build 2025 conference, Microsoft announced the launch of Windows AI Foundry, which rebrands and expands the Windows Copilot Runtime service introduced in May last year. Microsoft describes it as a "unified platform for local AI development"—a way to fine-tune, optimize, and deploy the underlying AI models for Windows applications.  

Windows AI Foundry can automatically detect the hardware of Windows machines and obtain the necessary software components to run specific AI models. Microsoft said that with the launch of new devices, Windows AI Foundry will also keep these components up to date and provide tools designed to simplify the model preparation process.  

Windows AI Foundry is a unified and reliable platform that supports the full AI developer lifecycle from model training to inference. Through simple visual and language task model APIs, developers can manage and run open-source large language models (LLMs) on Foundry Local or import proprietary models for conversion, fine-tuning, and deployment to clients and the cloud.  

### Edge Browser’s Local Model Capabilities Opened to Developers: PDF Translation Tool  

Microsoft launched new APIs for the Edge browser, allowing developers to integrate AI functions into web applications using models built into Edge.  

These new APIs give websites and Edge browser extensions access to the Phi 4 mini AI model, which has approximately 3.8 billion parameters. Microsoft says it performs well in mathematical problems, thanks to a training dataset that includes both manually created and synthetic data.  

Microsoft has introduced a PDF translation tool for Edge, supporting more than 70 languages. Users can simply open a PDF in Edge, click the "Translate" icon in the address bar, and create a new document translated into their chosen language. The PDF translation feature will be fully available next month.  

Edge also gained other AI APIs, including writing assistance APIs for generating, summarizing, and editing text. Microsoft plans to release a translation API in the coming months to provide AI-powered language translation through Edge.  

"These experimental APIs are designed to become potential web standards, working across platforms, browsers, and other AI models. For developers handling sensitive data or working in regulated industries, these APIs offer the privacy and security of device-local processing without sending data to external cloud services," Microsoft said.  

### NLWeb: Enabling Websites to Easily Build AI Chatbots  

Microsoft is launching an open project called NLWeb, aiming to simplify the process of building AI chatbots on the web. NLWeb allows websites to provide users with a "conversational interface" through a few lines of code, their chosen AI model, and their own data.  

"We believe [NLWeb] can play a role similar to HTML in the agentic web. It allows users to interact directly with web content in a rich, semantic way," Microsoft said.  

NLWeb enables websites to provide "conversational interfaces"—text fields and submit buttons—for users through a few lines of code, their chosen AI model, and their own data. For example, a retailer could use NLWeb to create a chatbot to help users select clothing for a specific trip, while a cooking website could use it to build a bot that suggests dishes to pair with recipes.  

Web pages built with NLWeb can optionally make their content discoverable and accessible to AI platforms that support MCP (Anthropic’s standard for connecting AI models with the systems where data resides).  

### The Arrival of "AI’s USB-C": Windows Integrates MCP Protocol  

With the official launch of the Azure AI Foundry Agent Service, Microsoft has brought new capabilities for professional developers, supporting the invocation of multiple specialized AI agents to handle complex tasks. This includes integrating Semantic Kernel and AutoGen to create a single developer-centric SDK, as well as supporting Agent-to-Agent (A2A) and Model Context Protocol (MCP).  

The open-source standard MCP, introduced by Anthropic last year and dubbed the "USB-C port of AI," was announced by Microsoft at this conference to be integrated into the Windows system. This protocol will enable AI applications and agents to communicate with other applications, web services, and even the Windows system itself in unprecedented ways.  

"As a platform, we want Windows to evolve so that agents become part of the operating system’s workload and a way for customers to continuously interact with applications and devices," Pavan Davuluri, Windows主管 (Head of Windows), told the media.  

Davuluri explained, "Agents can discover MCP servers installed on client devices through Windows’ MCP registry, leverage their expertise, and provide meaningful value to end users." These MCP servers will be able to access functions such as the Windows file system, window management, or the Windows Subsystem for Linux.  

Microsoft also announced new features in Azure AI Foundry Observability, including built-in metric monitoring tools covering key aspects like performance, quality, cost, and security, with detailed trace records integrated into a concise dashboard.  

### Open-Source Initiatives: New Command-Line Text Editor and More Tools  

At the Build 2025 conference, Microsoft open-sourced multiple applications and tools, including a new Windows command-line text editor called Edit.  

Edit will be pre-installed by default on Windows through the Windows Insider program this summer, allowing developers to edit files directly in the command line by running the "edit" command. Microsoft says the goal is to minimize unnecessary context switching and keep developers in their current workflow.  

Additionally, Microsoft will open-source GitHub Copilot in VS Code (its assistive coding extension for Visual Studio Code) and the Windows Subsystem for Linux (WSL), allowing developers to access the source code and make necessary changes for specific use cases.  

**Disclaimer**: The views in this article represent only the author’s personal opinions and do not constitute investment advice on this platform. This platform makes no guarantees regarding the accuracy, completeness, originality, or timeliness of the article’s information, nor shall it be liable for any losses arising from the use of or reliance on the article’s content.