Microsoft’s top 5 AI releases from Build 2025

The Rundown: Microsoft the GitHub Copilot coding agent, marking the evolution of Copilot from an AI assistant to an autonomous team member that can be assigned GitHub issues and create pull requests.

The details:

• The agent starts work when assigned a GitHub issue, creating a draft pull request and iterating based on review comments.

• It operates asynchronously by spinning up a secure development environment, and analyzing code using advanced reasoning.

• Available to Copilot Enterprise and Copilot Pro+ customers, it excels at tasks like adding features, fixing bugs, refactoring code, and improving documentation.

• Security is built-in: the agent respects branch protections, requires human approval before running CI/CD workflows, and follows custom security policies.

Why it matters: With the recent rise of AI coding agents like GitHub Copilot’s new coding agent, there’s a fundamental shift in how software gets built. Developers are transitioning from writing every line of code to becoming orchestrators of agents, delegating tasks while focusing on architecture, strategy, and creative problem-solving.

Image source: The Rundown / Kiki Wu

The Rundown: Microsoft is advancing its Windows AI strategy with for Model Context Protocol (MCP) on Windows 11 and the introduction of Windows AI Foundry — a foundation for AI agents to operate within the Windows ecosystem.

The details:

• MCP integration will bring Anthropic’s protocol to Windows 11, enabling AI agents to connect with native apps, system services, and external tools.

• Microsoft also the Windows AI Foundry, a new framework to help developers fine-tune and run AI models directly on Windows PCs. 

• Windows AI Foundry supports open-source and custom model deployment across CPUs, GPUs, and NPUs in Copilot+ PCs, enabling on-device capabilities.

Why it matters: Microsoft is pushing for Windows to be the premier platform for AI agent development and deployment. By moving AI processing to the client-side, the company is enabling faster, more secure, and privacy-conscious AI experiences.

Image source: The Rundown / Kiki Wu

The Rundown: Microsoft also Microsoft 365 Copilot Tuning, a low-code tool built into Microsoft Copilot Studio that enables orgs to fine-tune AI models using their own internal data, workflows, and domain expertise without requiring technical skills.

The details:

• Companies can train AI models on their proprietary documents, processes, and institutional knowledge to create company-specific agents in Agent Builder.

• The low-code tooling lets you build domain-specific AI agents that reflect organizational language, terminology, and format for targeted tasks.

• Copilot Tuning will roll out with three pre-built “recipes” that target tasks including Expert Q&A, Document Generation and Document Summarization.

Why it matters: With Copilot Tuning, Microsoft is making it easier for orgs to build tailored agents. By enabling companies to create agents from proprietary data without technical expertise, Microsoft is attempting to democratize customization previously limited to teams with significant engineering and data science resources.

Image source: The Rundown / Kiki Wu

The Rundown: Azure AI Foundry key updates including new AI models, fine-tuning, enhanced interoperability, and multi-agent orchestration that expand developers’ ability to design, customize, and manage AI apps and agents.

The details:

• The platform now offers access to , Black Forest Labs’ Flux Pro 1.1, alongside over 10K open-source models from .

• Developers can fully customize these models through fine-tuning techniques including LoRA/QLoRA and DPO, tailoring them for specific business use cases.

• Foundry Agent Service is now generally available, offering developers ready-to-use templates, actions, and connectors to build secure AI agents.

• Other tools include the for ranking top AI models by task and a , which selects the best model for each query in real time.

Why it matters: Microsoft’s Azure AI Foundry updates make it easier for developers to build and manage AI applications and agents that collaborate across complex workflows. These advances pave the way for more scalable, enterprise-ready AI solutions that can more seamlessly integrate with existing business data and processes.

Image source: The Rundown / Kiki Wu

The Rundown: At Build 2025, Microsoft also Microsoft Discovery, an AI-powered platform designed to revolutionize scientific R&D by deploying specialized AI agents throughout the entire research lifecycle.

The details:

• Microsoft Discovery taps specialized AI agents to automate and enhance every phase of the scientific research lifecycle, from ideation to experimentation.

• Built as a flexible, modular environment, it allows organizations to customize and extend workflows with industry-specific tools, plugins, and data sources.

• The platform fosters seamless collaboration between researchers and AI agents, enabling agents to handle the routine, data-intensive tasks.

• It’s built on top of a graph-based knowledge engine that maps complex relationships between proprietary data and scientific research.

Why it matters: Microsoft Discovery is a bold bet on AI-accelerated science. While the platform could completely revolutionize scientific research, its success hinges on whether AI can move beyond automating routine tasks to actually driving the creative problem-solving that leads to genuine scientific breakthroughs.

Image source: The Rundown / Kiki Wu

The Rundown: Speaking to journalists before Build and again on stage, Microsoft CTO Kevin Scott the company’s vision for an “agentic web” where AI agents move autonomously across platforms and tools in collaboration with people.

The details:

• Scott cautioned that closed approaches could stifle AI innovation, comparing it to how proprietary web protocols would’ve led to a “less interesting version of the web.”

• Microsoft is adding native support for open agent protocols including MCP and A2A across platforms like Copilot Studio and Azure AI Foundry.

• Scott was particularly excited about MCP (developed by Anthropic), comparing it to HTTP for its simplicity in connecting AI models with external tools.

• Microsoft also introduced , an open-source project to help turn websites and APIs into agentic apps, and make content accessible to agents using MCP.

Why it matters: Microsoft’s embrace of open AI protocols is a strategic shift from its historically closed approach, likely learning from past antitrust battles and browser market losses. This approach could ultimately determine whether we get a vibrant, interconnected agentic web or a landscape controlled by competing tech giants.