Unleashing Multi-Agent Power with the Agent2Agent (A2A) Protocol
During the last year, AI agents have rapidly advanced from being mere experimental projects to playing a crucial role in business solutions. What began as basic chatbots that simply replied to prompts, has now transformed into smart agents capable of making independent decisions and taking action for you. This transformation paints a new picture for software, where intelligence is no longer fixed within a single app or user interface.
Here at Microsoft, we’ve witnessed this change up close. More than 70,000 companies—ranging from start-ups to global enterprises like Atomicwork, Epic, Fujitsu, Gainsight, H&R Block, and LG Electronics—now use Azure AI Foundry to develop, tailor, and oversee AI applications and intelligent agents. In a short span of four months, over 10,000 organisations have started using our Agent Service for building and scaling networks of agents. Furthermore, Microsoft Copilot Studio has already served upwards of 230,000 organisations—including 90% of the Fortune 500.
As AI agents become increasingly capable, it’s essential they connect to a wide range of models, tools, and even each other. That’s why we are moving forward with open standards such as Agent2Agent (A2A), which will soon be available via Azure AI Foundry and Copilot Studio. This open protocol enables agents to collaborate smoothly across various platforms, clouds, and even between different organisations.
We’re committed to working with the wider industry to promote interoperability. Our approach continues to be focused on openness, giving practical support to developers, and turning experimental technology into robust, scalable solutions. Our mission is clear: empower every developer, whether professional or citizen, to build agents that can work together across every cloud and platform.
Microsoft Copilot has been designed to serve as the “AI interface” for staff everywhere, helping connect people to intelligent systems—where multiple agents can think, act, and adapt across multiple platforms without restriction. As organisations expand their agent networks, cross-system interoperability becomes vital. Customers increasingly require agents to manage tasks that span several vendors, cloud providers, and separate data stores. They also demand full visibility, control, and transparency—without being tied down to a single vendor.
That’s where protocols like A2A come in, supporting secure and structured agent communication. Agents can set goals, share updates, trigger actions, and securely exchange results—all while benefiting from advanced security measures like Microsoft Entra, mutual TLS, Azure AI Content Safety, and detailed audit logs. Whether using Semantic Kernel, LangChain, or other tools, developers can ensure connectivity and compliance as they build complex, distributed systems with trust and safety at the core.
Our latest developments
With the introduction of A2A support:
- Azure AI Foundry customers can now design advanced, multi-agent operations that combine internal tools, third-party services, and production assets—all with reliable management and oversight.
- Copilot Studio agents will be able to safely call on external agents, even those running outside the Microsoft ecosystem or made with different platforms.
- Businesses can assemble intelligent systems that adapt and scale across departments and cloud environments with ease.
- Microsoft’s support for A2A will help accelerate the industry-wide adoption of this open protocol.
This is merely the next step on our ongoing journey. Building on solutions like Autogen, the Semantic Kernel, our work on the Model Context Protocol (MCP), and our range of open models, we will continue to improve the platform with support for cutting-edge protocols, tools, and frameworks that matter most to both enterprise and developer communities. Protocols such as A2A and MCP bring us closer to achieving a truly connected agentic future.
Looking ahead
This wave of “agentic computing” isn’t just another tech fad—it’s a fundamental change to how technology is developed, how decisions are taken, and how value is created in business.
We have joined the A2A workgroup on GitHub to help shape the future of the protocol and supporting tools. A2A will be available as a public preview soon in Foundry and Copilot Studio.
As we support A2A and enhance our open orchestration platform, we are setting the stage for the next era of collaborative and adaptive software. The most effective agents will not be confined to one programme or environment—they will work seamlessly within your workflow, spanning multiple domains, platforms, and technologies. Our focus remains on openness, so that agents can freely cooperate, breaking down silos and delivering true intelligence throughout your organisation.
How to get started
If you’re eager to try out A2A, we’ve launched a new demonstration using the Semantic Kernel in Python. The sample shows how two local agents can work together to create a travel plan and carry out currency conversions—demonstrating straightforward interoperability without complex set-up or extra code.
