Microsoft has declared the “Age of AI Agents” at its Build 2025 conference on May 19, unveiling a comprehensive strategy and a suite of tools aimed at transforming software development and business automation. The company is championing an “open agentic web,” a concept where AI agents will autonomously interact, make decisions, and perform tasks for individuals and organizations, as detailed by Microsoft Chairman and CEO Satya Nadella. This vision underpins a significant push to make AI agents more capable, collaborative, and deeply integrated into daily workflows.
Key to this new era are substantial upgrades to GitHub Copilot, which evolves into a more autonomous coding assistant, and major expansions to Azure AI Foundry and Copilot Studio, designed for building, customizing, and orchestrating these intelligent agents. For developers and businesses, these announcements signal a shift towards more AI-driven assistance, potentially streamlining complex processes and unlocking new efficiencies. The new capabilities build upon Microsoft’s “Copilot Wave 2 Spring release,” which began introducing specialized agents and enhanced governance tools in April 2025.
The Developer Toolkit Reimagined
Microsoft is fundamentally altering the software development lifecycle, with the new GitHub Copilot coding agent at the forefront. Internally codenamed ‘Project Padawan’, this enhanced assistant is now generally available to Copilot Enterprise and Copilot Pro+ customers. As GitHub CEO Thomas Dohmke explained on the GitHub blog, the agent is directly embedded into the GitHub platform and can be assigned issues or prompted within VS Code.
It operates in a secure GitHub Actions-powered environment, autonomously handling tasks such as bug fixes, feature additions, and code refactoring. The Indian Express reported that it also excels at documentation improvements, particularly in well-tested codebases, and features plan-based execution before coding begins. Crucially, the agent pushes commits to draft pull requests, which then require human approval before any CI/CD (Continuous Integration/Continuous Deployment) workflows are initiated, ensuring a layer of oversight. Furthering its commitment to the developer community, Microsoft is also open-sourcing GitHub Copilot Chat in VS Code.
Azure AI Foundry, Microsoft’s unified platform for AI applications, now supports an expanded ecosystem of over 10,000 models, including new additions like Grok-3 and Grok-3 Mini from xAI. Microsoft also introduced tools like Model Router, in preview, to help select the optimal AI model for specific tasks.
The Azure AI Foundry Agent Service, now generally available, empowers developers to orchestrate multiple specialized agents for complex operations. Microsoft detailed here that this service integrates AI frameworks like Semantic Kernel and AutoGen into a unified SDK and supports Agent-to-Agent (A2A) communication and the Model Context Protocol (MCP).
Security is bolstered by Microsoft Entra Agent ID, in public preview, which automatically assigns unique identities to agents created in Copilot Studio or Azure AI Foundry. Additionally, Azure AI Foundry Local, in preview, brings on-device AI capabilities to Windows 11 and macOS.
Microsoft also launched Windows AI Foundry, an evolution of the Windows Copilot Runtime. It covers the entire development lifecycle, from model selection and optimization to deployment, integrating Foundry Local, Ollama (a platform for running large language models locally), and NVIDIA NIMs for access to open-source models across diverse Windows hardware. It also introduces ready-to-use AI APIs for Copilot+ PCs.
The company also announced Microsoft Discovery, an extensible platform to empower researchers with agentic AI, and a new open-source text editor for the command line called Edit.
AI Agents for Enhanced Business Processes
Microsoft is significantly extending AI agent capabilities into business operations through Microsoft 365 Copilot and Copilot Studio. A pivotal announcement is Microsoft 365 Copilot Tuning, a new low-code feature in Copilot Studio. This allows organizations to customize AI models using their own company data, workflows, and processes without requiring extensive data science expertise. For instance, a consulting firm could tune agents for specific industries based on its subject-matter expertise.
Copilot Studio is also gaining multi-agent orchestration capabilities, currently in public preview, enabling different AI agents to collaborate and divide work based on their expertise. A new Agent Store within Copilot Studio will allow users to discover, pin, and deploy agents from Microsoft, third-party partners like Jira and Miro, and custom agents built by their own companies.
Microsoft shared that over 230,000 organizations, including 90% of the Fortune 500, have already utilized Copilot Studio, with more than one million custom agents created across SharePoint and Copilot Studio in the last quarter alone. The “Copilot Wave 2 spring release,” featuring an updated Microsoft 365 Copilot app, a new Create experience, and the general availability of Copilot Notebooks, has also commenced its rollout. Specialized reasoning agents, Researcher and Analyst, are being deployed through the Frontier early access program.
Towards an Open Agentic Web and Navigating Challenges
Microsoft’s broader strategy involves fostering an “open agentic web.” The company is delivering broad first-party support for the Model Context Protocol (MCP) across its platforms, including GitHub, Copilot Studio, and Windows 11, and has joined the MCP Steering Committee.
MCP aims to standardize secure data access and interaction for AI agents. A new open-source project, NLWeb, was also introduced. Satya Nadella described this initiative as being akin to an “HTML for the agentic web.” Conceived by R.V. Guha, NLWeb is designed to make it easy for websites to offer conversational interfaces and function as MCP servers, with current sponsors including O’Reilly Media and Snowflake.
These advancements occur as the AI landscape rapidly evolves. Partners like OpenAI are also expanding connectivity for their models to enterprise data sources, such as GitHub, and Microsoft OneDrive and SharePoint. However, the rise of powerful AI agents is not without its challenges. Cybersecurity firm Pen Test Partners recently highlighted significant security vulnerabilities in Microsoft’s Copilot for SharePoint.
Their research demonstrated that AI agents could be manipulated through deceptive prompting to reveal sensitive data, bypass security measures, and potentially evade standard “accessed by” detection logs. This underscores a broader industry concern, with Gartner predicting that AI agent abuse will be behind “25% of enterprise breaches” by 2028, stemming from both external and malicious internal actors. Microsoft aims to mitigate such risks through initiatives like the Copilot Control System (CCS) and the new Microsoft Entra Agent ID.