Microsoft’s Build 2025 conference concluded this week. Observers noted that the event was much more developer-focused than usual. While there were some major announcements, the general sentiment was that there were not as many as in some previous years. Among the technical sessions and platform updates, two initiatives, NLWeb and Microsoft Discovery, drew particular attention for their potential applications.
NLWeb: A New Approach to Web Interaction
Microsoft introduced NLWeb (Natural Language Web), an open project aimed at simplifying the creation of natural language interfaces for websites. The goal is to allow users to query website content directly using natural language, similar to interacting with an AI assistant.
NLWeb utilizes semi-structured data already published by websites, such as Schema.org and RSS feeds, combining this with Large Language Model (LLM) powered tools. The system can also incorporate external knowledge from underlying LLMs to enhance query results, for example, by adding geographic information to a restaurant query.
Each NLWeb instance also functions as a Model Context Protocol (MCP) server, enabling websites to make their content discoverable to AI agents within the MCP ecosystem, should they choose to participate. As an open project, NLWeb is designed to be technology-agnostic, supporting various operating systems, AI models, and vector databases. Microsoft stated its belief that NLWeb could play a role similar to HTML in the context of an emerging “agentic web.” The project was conceived by R.V. Guha, who recently joined Microsoft and is known for his work on web standards like RSS and Schema.org.
The potential impact of NLWeb is considerable. NLWeb sounds cool, and it’s plausible that if enough sites integrate it, it could reduce the need for traditional web scraping by allowing more direct, conversational access to website information. The NLWeb GitHub repository now contains code and documentation for developers to begin working with the project.
Microsoft Discovery: AI Applied to Scientific Research
Another significant presentation involved Microsoft Discovery, a platform applying AI tools to scientific workflows. John Link, Chemistry Product Lead for Microsoft’s Science Platform, demonstrated how the platform was used in an effort to discover a PFAS-free alternative for immersion coolants, which are used in datacenters. Current coolants often use PFAS, sometimes referred to as “forever chemicals.”
The process involved AI agents conducting research over scientific knowledge, generating hypotheses for new chemical compounds, and running high-performance computing (HPC) simulations to validate candidates. Link stated, “This is not just a demo. We really did this,” emphasizing that the presented workflow represented an actual research project.
According to Microsoft, the digital discovery phase using AI models and HPC took approximately 200 hours. Following this, one of the identified candidates was synthesized in a lab in under four months. Tests reportedly showed that the material’s properties aligned with the AI predictions. A video demonstration showed a standard PC, running a demanding game, immersed in the synthesized coolant, maintaining stable temperature without fans.
The capabilities shown have led to some positive reactions. Many can see Microsoft Discovery leading to a new wave of scientific innovation, and that prospect is considered cool. There is also expressed hope that the specific coolant invented by Microsoft eventually comes to market.
Other Notable Announcements
Reflecting the developer focus of the event, Microsoft announced a range of other updates:
- Microsoft Copilot Studio and Microsoft 365 Copilot: Updates included new pro-code capabilities for Copilot Studio, such as the generally available Microsoft 365 Agents Toolkit for Visual Studio. The Microsoft 365 Copilot Wave 2 spring release is rolling out with an updated app and OpenAI GPT-4o image generation.
- Azure AI Foundry: The platform is expanding with new models, including Grok 3 from xAI and upcoming availability of Flux Pro 1.1 and Sora (via Azure OpenAI). It now offers access to over 10,000 open-source models from Hugging Face and will make Azure AI Foundry Local available on Windows 11 and MacOS.
- Windows AI Foundry: Introduced as a unified platform for local AI development on Windows, it will integrate Foundry Local and provide access to open-source models.
- Data and Developer Tools: Microsoft SQL Server 2025 will include built-in AI capabilities. Agentic DevOps concepts are being introduced to GitHub Copilot.
- Platform Updates: New AI APIs are coming to Microsoft Edge. The Microsoft Power Platform received updates for AI agent collaboration. The Microsoft Store on Windows will introduce zero onboarding fees for individual developers starting in June 2025 (preview).
- Security: Microsoft Entra Agent ID is now in preview, and the company is contributing to identity and registry standards for the MCP ecosystem. Post-quantum cryptography algorithms are in preview for Windows Insiders and Linux.
These announcements indicate Microsoft’s continued investment in AI across its product lines and developer tools.