A2A Strategy Core: Dual Definitions and the Cornerstone of "Agentic Computing"
Microsoft's "A2A" strategy, simply put, encompasses two main aspects: Firstly, the familiar "Application-to-Application" automation, which helps different software systems communicate smoothly and break down information silos. Secondly, the future-oriented "Agent-to-Agent" collaboration, enabling AI agents to communicate and cooperate seamlessly across platforms and organizations. This is not just a technological trend, but a fundamental shift towards "Agentic Computing".
Microsoft's goal is to build a new ecosystem of highly interconnected and intelligently autonomous enterprise systems. By embracing open protocols like A2A and integrating them deeply, Microsoft is actively promoting interoperability in the AI field, committed to creating an open, collaborative intelligent future where intelligence is no longer confined to a single application or interface.
Two Strategic Pillars: Unveiling the Technological Foundation
Agent-to-Agent (A2A) Protocol
Microsoft adopts the open A2A standard (originating from Google), allowing AI agents to communicate without barriers, even if they are built with different tools like Semantic Kernel or LangChain. Key technical points:
- Agent Cards: The "business card" of an agent, providing a quick understanding of its capabilities.
- Tasks & Lifecycles: Standardized task flows for easy management and tracking.
- Messages & Artifacts: Clear communication formats for efficient exchange of information and results.
- Streaming & Push: Real-time updates and proactive notifications for more timely interactions.
The A2A protocol enables structured agent communication—securely and observably exchanging goals, managing states, invoking actions, and returning results.
Enterprise-Grade Security: Microsoft Entra ID mTLS Mutual Encryption Azure AI Content Safety and complete audit logs ensure secure and trustworthy communication.
Azure Integration Services (AIS)
The "Swiss Army knife" for enterprise application connectivity, providing a solid backbone for traditional application integration and emerging agent scenarios. Core services include:
- Logic Apps: Build automated business processes like assembling building blocks.
- Service Bus: A reliable message "courier" ensuring secure message delivery.
- API Management: The "steward" of APIs, unifying publishing, management, and protection.
- Event Grid: An event "broadcasting station" for real-time responses to various system events.
- Data Factory: A data "processing plant" for efficient integration and transformation of massive data.
AIS is the "superhighway" for AI agents to interact smoothly with existing enterprise systems, ensuring agents can effectively access data and execute actions.
Core Enabling Platforms: Accelerate Your Agent Application Development
Azure AI Foundry
Trusted by over 70,000+ enterprises and digital-native companies worldwide (such as Atomicwork, Epic, Fujitsu, Gainsight, H&R Block, LG Electronics), it's an "agent factory". The new Agent Service attracted 10,000+ organizations in just 4 months. Through the A2A protocol, you can build and orchestrate complex, multi-agent workflows spanning internal and external tools, while ensuring governance and Service Level Agreements (SLAs).
Microsoft Copilot Studio
Over 230,000 organizations (including 90% of Fortune 500 companies) are already using it. With A2A interoperability, agents built in Copilot Studio will be able to securely call external agents, even those built on other platforms or hosted outside the Microsoft ecosystem, greatly expanding Copilot's capabilities.
Semantic Kernel
An open-source "magic wand" that easily combines large language models with programming languages like C# and Python to build AI agents capable of A2A communication. Microsoft is committed to supporting the tools developers know and love.
What Unique Value Can the A2A Strategy Bring to Enterprises?
Skyrocketing Operational Efficiency
Automate complex processes, enable agent collaboration, and unleash employee potential.
Data-Driven Decisions
Improve data consistency and accuracy for deeper insights and smarter decisions.
IT Architecture Modernization
Enhance system scalability and future adaptability, fearless of technological iterations.
AI-Powered Innovation
Enable proactive operations, accurate predictions, and foster new intelligent applications.
Build Composable Intelligent Systems
Flexibly assemble intelligent capabilities across organizational and cloud boundaries to build solutions on demand.
Open and Interconnected Ecosystem
Break down technological barriers, promote cross-vendor and cross-platform collaboration, and avoid lock-in.
Embracing A2A: Challenges and Security Issues to Consider
Main Challenges
- Integration of new and old systems: Implementation can be somewhat complex.
- Cost and resources: Requires careful planning for investment and management.
- Evolution of technical standards: Keep learning and stay updated.
AI Agent Security Risks
- "Tricking" agents: Beware of prompt injection attacks.
- Tools being "corrupted": Prevent abuse of external tools.
- Identity "impersonation": Guard against spoofing and communication poisoning.
- Code "being executed": Mitigate remote code execution risks.
How Microsoft Responds: Security and Trust are Foundational
Microsoft deeply integrates A2A communication into its mature enterprise-grade security framework (e.g., Microsoft Entra ID for authentication, mTLS for channel encryption, Azure AI Content Safety for content moderation), complemented by sandboxed execution, least privilege, Data Loss Prevention (DLP), and comprehensive audit logs, actively building a trustworthy, secure, and compliant agent ecosystem. Azure AI Foundry has trust mechanisms built-in by default, ensuring that even in an increasingly open and distributed environment, security, compliance, and accountability remain paramount.
The Future is Here: "Agentic Computing" Leads a New Software Paradigm
"Agentic Computing" is not a fleeting trend, but a fundamental shift in how software is built, decisions are made, and value is created. Microsoft's A2A strategy is a key cornerstone of this grand vision.
Microsoft is committed to promoting open standards and has joined the A2A GitHub working group to contribute specifications and tools. A public preview of the A2A protocol will soon be available in Azure AI Foundry and Copilot Studio. Microsoft will continue to invest in Autogen, Semantic Kernel, Model Context Protocol (MCP), and open models to support the protocols, models, and frameworks most needed by developers and enterprises.
The next generation of software will be:
The best agents will not exist in isolation; they will operate seamlessly within workflows, spanning models, domains, and ecosystems. Microsoft is building this future with openness at its core, because intelligence should be as boundless and collaborative as the world it serves.
Hands-on: Semantic Kernel and A2A Protocol Integration Example
Learn how to achieve instant, secure, and asynchronous interoperation between cross-cloud agents using the lightweight JSON-RPC A2A protocol with Azure AI Foundry's Semantic Kernel framework, without exchanging code or credentials.
Microsoft's Contribution to the A2A Ecosystem
The A2A protocol is currently in its early stages of development and lacks packaged libraries, posing challenges for developers to quickly build integrations. Microsoft's initial contribution aims to demonstrate how Semantic Kernel agents can effectively integrate into the A2A ecosystem by directly leveraging existing sample code from the A2A codebase. The goal is to provide a clear integration path, enabling customers to easily adopt Semantic Kernel in their A2A projects.
SemanticKernelTravelManager
Acts as a central coordinator, receiving and analyzing requests, and intelligently routing tasks (like currency queries, itinerary planning) to specialized agents based on context.
CurrencyExchangeAgent
Handles currency-related tasks, integrating external APIs (like Frankfurter API) to provide real-time exchange rates, aiding in accurate budgeting and financial planning.
ActivityPlannerAgent
Provides personalized itinerary suggestions, activity, and event bookings based on user preferences and budget, creating customized travel experiences.
How Does the Integration Work?
- Task Routing and Delegation: The TravelManager (as coordinator) configures its specialized agents as plugins. Using context-awareness and automatic function calling, the underlying model intelligently determines the most appropriate agent to handle the request.
- Agent Discovery: Agents declare their capabilities via structured "Agent Cards," enabling client agents to efficiently identify and select the most suitable agent for a specific task and communicate seamlessly via the A2A protocol. This example declares a Semantic Kernel agent.
- Conversation Memory: Semantic Kernel maintains context across multi-turn interactions through its chat history, providing a smooth user experience (the history in this example is ephemeral and not persisted).
Example Scenario: Plan Your Trip with Ease
Imagine a user wants to plan a budget-friendly trip and needs currency conversion:
- User submits a request to
TravelManager. TravelManagerdetects the currency conversion need and callsCurrencyExchangeAgent.CurrencyExchangeAgentfetches exchange rates from the Frankfurter API.ActivityPlannerAgentsuggests budget-friendly activity options based on the budget.TravelManageraggregates the information and returns a complete travel plan.
Quick Start Guide
Ensure you have locally cloned the latest version of the A2A codebase to successfully run this demo.
1. Prerequisites:
- Python 3.10 or higher
uvpackage manager (see uv documentation for installation instructions)- Valid OpenAI credentials (see SK documentation)
- Optional: Frankfurter API key (free endpoint does not require a key)
2. Setup and Run:
Navigate to the Semantic Kernel example within the A2A samples directory:
cd samples/python/agents/semantickernelCreate a .env file and fill in your credentials:
OPENAI_API_KEY="your_api_key_here"
OPENAI_CHAT_MODEL_ID="your-model-id"Set up the environment (pin to your desired/installed Python version):
uv python pin 3.12 # or your Python version
uv venv
source .venv/bin/activate # Linux/macOS
# .venv\Scripts\activate # WindowsRun the Semantic Kernel Agent:
# Default configuration
uv run .
# Or run with custom host and/or port
uv run . --host 0.0.0.0 --port 8080After running, you will see output indicating the server has started.
In another terminal, run the A2A client:
cd samples/python # Go back to the python directory of A2A samples
uv run hosts/cli --agent http://localhost:10020 # Assuming SK Agent is running on port 10020After running the client, you will see the Agent Card displayed, confirming the Semantic Kernel Travel Agent Manager has been successfully discovered and is ready for interaction.
Now you can send queries directly to the agent, for example:
I am traveling to Seoul, South Korea for 2 days.
I have a budget of $100 USD a day.
How much is that in South Korean Won? What sort of things can I do and eat?After submitting the request, you will see streaming event messages in the console reflecting the different stages of request processing, eventually receiving a structured JSON response.
Next Steps
Microsoft is actively developing more integration examples, such as connecting to Azure AI Foundry via Semantic Kernel's AzureAIAgent.
Additionally, we plan to contribute an A2A example directly to the Semantic Kernel codebase. Stay tuned for future updates!
How to Successfully Implement the A2A Strategy and Maximize Value?
Identify Pain Points: Start with business scenarios that can yield high returns and solve real problems.
Assess Current State: Understand your existing IT systems, data, network, and security posture to prepare.
Security First: Plan AI security and governance strategies from the outset, leveraging Microsoft security tools.
Develop Talent: Invest in team skills to master AI, A2A, and Azure-related knowledge.
Start Small, Iterate Fast: Begin with pilot projects, validate success, then gradually roll out, iterating agilely.
Encourage Innovation: Foster an atmosphere to explore A2A application opportunities across various businesses and discover new value.
Stay Informed: Keep up with the latest developments in A2A protocols, MCP, and other relevant standards, adjusting strategies accordingly.
Leverage Tools: Actively use Microsoft platforms and tools like Azure AI Foundry, Copilot Studio, and Semantic Kernel to accelerate development.
Core Idea: Don't just think about how to use A2A to make existing processes faster; more importantly, consider "What new business models and processes become possible with A2A?" Embrace the concept of a "composable enterprise," flexibly building future business capabilities with specialized AI agents and integrated applications.