The Agent2Agent Protocol (A2A) is being announced.
A new era of Agent Interoperability
AI agents present a special chance to increase human productivity by managing a large number of repetitive or complicated daily chores on their own. In order to help scale, automate, and improve workplace procedures from helping customer service agents to helping with supply chain planning to acquiring new laptops businesses are increasingly creating and implementing autonomous agents.
These agents must be able to cooperate in a dynamic, multi-agent ecosystem across fragmented data systems and applications in order to optimise the advantages of agentic AI. Enabling agents to interact together, even if they were developed in separate frameworks or by different vendors, will reduce long-term costs, boost autonomy, and compound productivity gains.
With the help of more than 50 technology partners, including Atlassian, Box, Cohere, Intuit, Langchain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, UKG, and Workday, as well as top service providers, including Accenture, BCG, Capgemini, Cognisant, Deloitte, HCLTech, Infosys, KPMG, McKinsey, PwC, TCS, and Wipro, one can are launching a new, open protocol today called Agent2Agent (A2A).
On top of different enterprise platforms or apps, the A2A protocol will enable AI agents to coordinate operations, securely exchange information, and communicate with one another. Since clients’ AI agents may now operate throughout their whole enterprise application estate, believe the A2A framework will be very beneficial to them.
This cooperative endeavour represents a common vision of a time where AI agents can work together seamlessly to automate intricate enterprise procedures and spur previously unheard-of levels of creativity and efficiency, irrespective of the underlying technologies.
Anthropic’s Model Context Protocol (MCP), which gives agents useful tools and context, is enhanced by the open protocol A2A. That created the A2A protocol to solve the issues encountered when implementing large-scale, multi-agent systems for customers, by leveraging Google’s in-house experience in scaling agentic systems.
A2A gives consumers the freedom to mix agents from different sources and enables developers to create agents that can connect with any other agent created using the protocol. Most importantly, companies gain from a standardized approach to agent management across various cloud environments and platforms. For collaborative AI agents to reach their full potential, they believe this universal interoperability is crucial.
A2A design principles
Regardless of the underlying framework or vendor, A2A is an open protocol that offers a uniform method for agents to operate together. It used five important guidelines when working with others to build the protocol:
- Embrace agentic capabilities: A2A aims to allow agents to work together in their organic, unstructured modalities, even in situations where they are not sharing context, memory, or tools. Without restricting an agent to a “tool,” we are enabling real multi-agent scenarios.
- Build on existing standards: Because the protocol is based on well-known, established standards like HTTP, SSE, and JSON-RPC, it is simpler to integrate into the IT stacks that companies currently use on a regular basis.
- Secure by default: At launch, A2A’s authentication schemes are equal to those of OpenAPI, and it is built to handle enterprise-grade authorization and authentication.
- Support for lengthy work: Everyone built A2A to be adaptable and handle situations in which it performs exceptionally well, from short jobs to in-depth research that could take hours or even days when people are involved. A2A can give its users real-time feedback, notifications, and state updates during this procedure.
- Modality agnostic: Because the agentic world isn’t simply text, so built A2A to accommodate a variety of modalities, such as streaming audio and video.
How A2A works
A2A makes it easier for a “client” agent and a “remote” agent to communicate. While the remote agent is in charge of carrying out the tasks in an effort to deliver the right information or take the right action, the client agent is in charge of creating and transmitting the tasks. Several essential skills are involved in this interaction:
- Capability discovery: Agents can use a JSON-formatted “Agent Card” to advertise their capabilities. This enables the client agent to choose the most qualified agent for a job and use A2A to connect with the remote agent.
- Task and state management: In order to complete tasks and satisfy end-user requirements, remote agents and clients communicate with one other. The protocol defines this “task” object, which has a lifecycle. It can be finished right away, or for lengthy projects, each agent can communicate with the others to keep each other updated on the most recent progress of a work. An “artefact” is the result of a task.
- Secure Collaboration: To exchange context, responses, artefacts, or user instructions, agents can message one another.
- User experience negotiation: Every message contains “parts,” which are whole pieces of material, such as a produced image. Because each component has a designated content type, the client and remote agents can explicitly negotiate the user’s UI capabilities, such as iframes, video, web forms, and more, as well as the appropriate format that is required.
A real-world example: candidate sourcing
With A2A collaboration, hiring a software engineer can be made more easier. A user (such as a hiring manager) can assign an agent the responsibility of finding applicants that fit a job posting, location, and skill set inside a centralized interface like Agents pace. To find possible applicants, the agent then communicates with other specialised agents.
After receiving these recommendations, the user can instruct their agent to set up additional interviews, expediting the process of finding candidates. Background checks might be facilitated by another agent once the interview procedure is finished. This is only one illustration of how Artificial Intelligence agents must work together across systems in order to find a suitable job applicant.
The future of agent interoperability
A2A could usher in a new era of agent interoperability, encouraging creativity and producing agentic systems that are more potent and adaptable. This protocol, in the hope that it will open the door to a day when agents may work together harmoniously to improve their lives and address challenging issues.
