Build 2025: Microsoft opens up Windows machine learning

Microsoft’s annual developer conference, Build 2025, has given a strong indication of how artificial intelligence (AI) is being fully integrated into the Windows operating system. Windows developers are being given access to the hardware-agnostic AI engine that powers Copilot+ PCs and can tune the engine using their own training data, to optimise it when integrated into their own applications.

Pavan Davuluri, corporate vice-president of Windows + Devices at Microsoft, wrote in a blog post that Microsoft’s north star was to make Windows a great platform for developers, where AI is integrated into the software, hardware and silicon. 

During the opening day of Build 2025, the company unveiled the Windows AI Foundry, which it said offers a unified and reliable platform supporting the AI developer lifecycle from model selection, optimisation, fine-tuning and deployment across client and cloud.

Windows AI Foundry provides access to Windows ML, the AI inference engine in Windows via application programming interfaces (APIs) for language and vision tasks, such as text intelligence, image description, text recognition, custom prompt and object erase.

Windows ML is effectively a hardware-agnostic interface to underlying AI acceleration chipsets on PCs that uses the Onnx runtime engine. Analogous to the DirectX API that games developers use to access graphics processing units, Microsoft said it is working with hardware partners to ensure conformity and compatibility.

Adobe is one of the software companies looking at using the new AI APIs in Windows. “Adobe Premiere Pro and After Effects juggles terabytes of footage and heavy ML workloads,” said Volker Rölke, a senior machine learning computer scientist at Adobe.

“A reliable Windows ML API that delivers consistent performance across heterogeneous devices would remove huge obstacles and let us ship more exceptional features faster. Windows ML can help us take a hardware-agnostic approach with far less boiler-plate system checks and low-level decision-making.”

Microsoft is also providing low-rank-adaption functionality, which Davuluri said is used with custom data to fine tune Microsoft’s Phi Silica local language model, which is built into Copilot+ PCs.

The company is also providing APIs for semantic search and knowledge retrieval, which he said can be used by developers to build natural language search and retrieval-augmented generation into their Windows application.

Davuluri said Windows AI Foundry also enables developers to use their own models and deploy them on AMD, Intel, Nvidia and Qualcomm-based chipsets.

As it opens up the built-in AI to Windows developers, Microsoft said it’s also integrating Model Context Protocol (MCP) with Windows to provide a standard framework for AI agents to connect with native Windows apps. Davuluri said this enables apps to participate seamlessly in agentic interactions. “Windows apps can expose specific functionality to augment the skills and capabilities of agents installed locally on a Windows PC,” he said.

In the Build 2025 opening keynote, Microsoft CEO Satya Nadella discussed the opportunity MCP provides in enabling agents to coordinate business flows.

“At some level, you can think about these agents and multi-agent frameworks orchestrating workflows in an agentic way for every role and every business process once every single business application starts showing up as an MCP server,” he said.

Nadella believes such capabilities will be a game changer for developers creating next-generation workflow and business process automation.