Over the previous 12 months, we’ve seen AI brokers evolve from experimental instruments to important elements of enterprise techniques. From easy immediate and response bots to brokers that act autonomously in your behalf, this shift marks a brand new period of software program design the place intelligence is now not tied to static interfaces or single functions.
At Microsoft, we’ve seen this transformation firsthand. Azure AI Foundry is now utilized by builders at greater than 70,000 enterprises and digital native corporations, together with Atomicwork, Epic, Fujitsu, Gainsight, H&R Block, and LG Electronics, to design, customise, and handle AI apps and brokers. In simply 4 months, over 10,000 organizations have adopted our new Agent Service to construct, deploy, and scale agentic techniques. Greater than 230,000 organizations, together with 90% of the Fortune 500, have already used Microsoft Copilot Studio.
As brokers tackle extra subtle roles, they want entry not solely to numerous fashions and instruments but in addition to at least one one other. That’s the reason we’re dedicated to advancing open protocols like Agent2Agent (A2A), coming quickly to Azure AI Foundry and Copilot Studio, which can allow brokers to collaborate throughout clouds, platforms, and organizational boundaries.
We’re aligning with the broader trade push for shared agent protocols—doing what we’ve at all times executed: embracing openness, supporting real-world builders, and turning experimentation into enterprise-grade platforms. Our objective is easy: empower each professional and citizen builders to construct brokers that interoperate throughout clouds and frameworks.
We consider that Microsoft Copilot will empower each worker and act because the “UI for AI” to attach with brokers and agentic techniques—networks of brokers that cause, act, and adapt throughout boundaries. As clients scale these techniques, interoperability is now not non-compulsory. They need their brokers to orchestrate duties that span distributors, clouds, and information silos. They need management, visibility, and belief—with out being locked in.
A2A can allow structured agent communication—exchanging objectives, managing state, invoking actions, and returning outcomes securely and observably. Builders can use instruments they know, like Semantic Kernel or LangChain, and nonetheless interoperate. Each name travels by enterprise-grade safeguards: Microsoft Entra, mutual TLS, Azure AI Content material Security, and full audit logs. Azure AI Foundry is constructed with belief by default, and as agent ecosystems develop extra open and distributed, security, compliance, and accountability stay first-class.
What we’re delivering
With help for A2A:
- Azure AI Foundry clients can construct advanced, multi-agent workflows that span inside copilots, associate instruments, and manufacturing infrastructure—whereas sustaining governance and SLAs.
- Copilot Studio brokers will be capable of securely invoke exterior brokers, together with these constructed with different platforms or hosted outdoors Microsoft.
- Enterprises acquire a path to composable, clever techniques that scale throughout organizational and cloud boundaries.
- Microsoft’s contributions will speed up growth and adoption of the open A2A protocol throughout the trade.
This is only one step on an extended journey. As we’ve executed with improvements like Autogen, Semantic Kernel, our contributions to Mannequin Context Protocol (MCP), and our catalog of open fashions, we’ll proceed to evolve the platform to help the protocols, fashions, and frameworks that matter most to builders and enterprises. We see protocols like A2A and MCP as essential steps within the path of realizing our imaginative and prescient for the agentic future.
What’s subsequent
Agentic computing isn’t a development—it’s a foundational shift. It adjustments how software program is constructed, how selections are made, and the way worth is created.
We’ve joined the A2A working group on GitHub to contribute to the spec and tooling. The A2A public preview in Foundry and Copilot Studio will arrive quickly.
By supporting A2A and constructing on our open orchestration platform, we’re laying the muse for the following technology of software program—collaborative, observable, and adaptive by design. The very best brokers gained’t stay in a single app or cloud; they’ll function within the circulation of labor, spanning fashions, domains, and ecosystems. We’re constructing that future with openness on the middle—as a result of brokers shouldn’t be islands, and intelligence ought to work throughout boundaries, identical to the world it serves.
Getting began
We’ve launched a new pattern in Semantic Kernel (obtainable in Python) that demonstrates how two native brokers can collaborate utilizing the A2A protocol. On this instance, the brokers work collectively to plan a journey itinerary and deal with foreign money conversions, showcasing seamless interoperability with out the necessity for customized orchestration code.