Top 10 Announcements from Microsoft Build 2025 - AI Agents Grow Up!
#40 | Agents grow up!
Microsoft Build is today, and if I could describe the key insights from the developer keynote, it would be that Agentic AI is Growing Up!
Disclosure: I work at Microsoft!
Well, what do I mean? Earlier this year, I wrote about how most agent deployments were primarily focused on calling tools, with very little autonomous multi-agent composition. The production story was still in its early stages, and had not gotten to the point of clearly addressing what tools customers could use to enable a set of use cases at scale.
This May, Microsoft Build keynote announcements (by Satya Nadella) changed all of that with products showing autonomous agentic capabilities (e.g., GitHub Agents, M365 Analyst, M365 Researcher), the Azure AI Foundry Agent Service that lets you build, deploy, and scale agents, and the Teams AI library that lets you build agents and publish them to a store where they can be reused across platforms like Teams and M365.
1. Improvements to GitHub: As part of a broader commitment to helping build the open agentic web
GitHub Copilot is evolving from an in-editor assistant to an agentic AI partner with a first-of-its-kind asynchronous coding agent. Starting today, GitHub Copilot Chat in VS Code is now opensource, making AI-powered capabilities part of the same repository driving the world's most popular development tool. This reinforces the clear commitment to open, collaborative, AI-powered software development.
Related resources:
2. New Preview Apps: SRE Agent - automatic proactive SRE work
Azure SRE Agent is an AI-powered tool that makes it easier to maintain production cloud environments. It continuously learns about and monitors Azure resources, responds to incidents automatically, and helps with rapid root cause analysis. Key capabilities include evaluating performance trends, proactively detecting security vulnerabilities, automating incident response, and performing mitigation actions like scaling resources or rolling back deployments. SRE Agent can reduce incident resolution time from hours to minutes and seamlessly integrates with observability tools and GitHub for closing the loop with developers.
Related resources:
3. OpenAI (Roadmap) - new models and improved usefulness
In terms OpenAI’s roadmpa, Sam Altman in his short segment at the keynote hinted at a future focus on models that are easier to use (e.g., requests routed automatically to the right model with the right capabilities including multimodality, tool use , computer use etc).
IMO, the biggest gains we will get for the rest of the year will be related to more and more capabilities being “lifted up” into the model and fine-tuned end to end for better performance (I wrote about this earlier this year).
4. Agents in Products: New agents in Microsoft 365 (Researcher, Analyst)
These apps were released earlier but worth mentioning. Researcher helps tackle complex, multi-step research tasks by combining OpenAI's deep research model with Microsoft 365 Copilot's advanced orchestration and deep search capabilities. It can securely access your work data and the web to develop. Analyst thinks like a skilled data scientist, using OpenAI's o3-mini reasoning model to perform advanced data analysis through chain-of-thought reasoning. It can run Python to tackle complex data queries and transform raw data into forecasts, visualizations, or projections in minutes.
Related resources:
5. Agent Store: You can now add agents in teams, mention them with a streamlined dev experience
Satya discussed early work on the Microsoft Agent Store. This enables developers to build and deploy agents across Teams and Microsoft 365 platforms with a streamlined development experience.
Related resources:
6. Build and Deployment at Scale: Azure Foundry - the full production ready agentic stack
Perhaps the most exciting announcement today.
How do I deploy my autogen agents? Use Azure AI Foundry Agent Service!
Think of Azure AI Foundry as the production stack for AI agents. From models provisioning, to governance, deployment, evaluation etc. Azure AI Foundry Agent Service is now generally available, bringing new capabilities to empower professional developers to orchestrate multiple specialized agents for complex tasks. This includes bringing Semantic Kernel and AutoGen into a single developer-focused SDK with Agent-to-Agent (A2A) and Model Context Protocol (MCP) support. New features in Azure AI Foundry Observability provide built-in metrics for performance, quality, cost and safety.
Related resources:
7. AI on the Edge: In Windows - Windows AI Foundry & Phi Silica
Windows AI Foundry offers a unified and reliable platform supporting the AI developer lifecycle across training and inference. With simple model APIs for vision and language tasks, developers can manage and run open source LLMs via Foundry Local or bring a proprietary model to convert, fine-tune and deploy across client and cloud.
Related resources:
8. Building the Open Agentic Web: Memory as a critical factor & NLWeb
Microsoft is introducing NLWeb, which will play a similar role to HTML for the agentic web. NLWeb makes it easy for websites to provide a conversational interface for their users with the model of their choice and their own data. Every NLWeb endpoint is also an MCP server, so websites can make their content easily discoverable and accessible to AI agents.
Related resources:
9. AI for Science: Microsoft Discovery
Microsoft Discovery is an extensible platform built to empower researchers to transform the entire discovery process with agentic AI. It helps research and development departments across various industries accelerate the time to market for new products and expand the end-to-end discovery process for all scientists.
Related resources: