Microsoft wants to redesign the web for intelligent agents. Here's what's new at Microsoft Build

With over 15 million developers already using GitHub Copilot and more than 230,000 organisations, including 90 per cent of the Fortune 500, leveraging Copilot Studio to create AI agents, Microsoft confirms its entry into the era of intelligent, autonomous agents. At its annual Build 2025 conference, the company announced several innovations that redefine the software development cycle and introduce open standards for building an agent web. The latter is the most interesting announcement because it suggests the scope of the intelligent agent era. The Redmond giant presented NLWeb, a new HTML-like standard that allows direct and semantic interactions between users, web content and AI agents. The idea seems to be to enable websites to offer conversational interfaces based on customised AI models and proprietary data, allowing users to interact with web content. More simply, websites can - if they wish - make their content queryable and accessible to AI agents. To make an agent economy and thus an AI economy a reality, the web also needs to conform. "We imagine a world in which agents operate at the individual, organisational, team level and along the entire chain of business processes," the note reads. "This emerging vision of the Internet is that of an open agent web, in which AI agents make decisions and perform tasks on behalf of users or organisations.

GitHub Copilot becomes an asynchronous coding agent integrated into the GitHub platform, now home to more than 150 million developers. Microsoft makes GitHub Copilot Chat open-source for VS Code and adds tools to manage prompts, assessments and business checks. The Windows AI Foundry platform is born. It is a dedicated environment for developers for the entire AI lifecycle, from training to inference, allowing them to manage open-source or proprietary language and vision models. Through simplified APIs for vision and language models, developers can manage and run open-source large language models (LLM) through Foundry Local, or use proprietary models to be converted, optimised and deployed either locally or in the cloud. Windows AI Foundry is already available to start developing.Next: With over 1,900 AI models available, Azure AI Foundry welcomes xAI's new Grok 3 and Grok 3 mini models, offering innovative tools such as Model Leaderboard and Model Router to optimise the use of AI resources.

At the Build conference, Microsoft introduces new pre-configured agents, modular custom-build components, multi-agent capabilities and new templates to help developers and organisations securely design and deploy agents, significantly improving productivity. Azure AI Foundry Agent Service introduces new capabilities that enable professional developers to orchestrate specialised agents capable of tackling complex tasks. The service integrates Semantic Kernel and AutoGen into a single developer-oriented SDK, with support for Agent-to-Agent (A2A) communication and Model Context Protocol (MCP). As a tribute to transparency, Azure AI Foundry Observability is christened a tool offering integrated metrics on performance, quality, cost and security, with detailed traceability accessible via a centralised dashboard.

Thanks to Microsoft Entra Agent ID (currently in preview), agents created with Microsoft Copilot Studio or Azure AI Foundry automatically receive a unique identity registered in the Entra directory, enabling companies to manage agents securely from the outset and avoid 'agent sprawl' (uncontrolled proliferation). Applications and agents developed in Foundry will also reportedly benefit from Microsoft Purview's security and compliance controls. Foundry also offers advanced governance tools to define risk parameters, initiate automated assessments and receive detailed reports.

Microsoft 365 Copilot Tuning is a feature for training templates and creating agents in a simplified mode, using low-code tools. These agents perform specialised tasks with high precision, operating securely within the boundaries of Microsoft 365 services. For example, a law firm can create an agent capable of generating documents consistent with the organisation's style and expertise. In addition, the new multi-agent orchestration feature in Copilot Studio promises to connect multiple agents together, enabling collaboration on larger and more complex tasks.