Where MCP focuses on communication between agents and external or internal systems, protocols like A2A enable agents to discover each other, interact, collaborate and delegate tasks, whether operating within an organization’s security perimeter or outside it. These protocols address a growing need in complex environments where multiple agents work together across organizational or technical boundaries, enabling agents from different vendors to communicate effectively.Released by Google in April 2025,7 A2A operates through a common communication interface and introduces the concept of agent cards (similar to model cards8), which are structured descriptions of an agent’s identity, along with its capabilities and skills. This allows for automatic discovery and coordination between agents and systems. Illustration of agent-to-agent communication protocol FIGURE 4 AI agent 1 AI agent 2 Agents LLM Agent frameworkAgents LLM Agent frameworkAgents card Task manager Artefact handlerA2A protocol MCP A2A APIs and enterprise applicationsAPIs and enterprise applications Beyond communication and discovery, new standards are also emerging that address how agents transact and exchange value. Released by Google in September 2025, the Agent Payments Protocol (AP2)9 enables secure, auditable transactions under user-defined limits. Unlike MCP and A2A, which focus on data exchange and task coordination, AP2 addresses complex financial operations. Despite this progress, interoperability remains a key challenge. Technical compatibility alone does not guarantee successful coordination between agents. Strategy, privacy and security considerations often shape how and whether systems should be integrated and are important for enterprises to carefully consider. For example, communication between different agents could raise concerns about access control, data confidentiality or compliance across jurisdictions. Choosing whether to expose a capability to other agents becomes a governance decision as much as a technical choice. AI Agents in Action: Foundations for Evaluation and Governance 10

AI Agents in Action Foundations for Evaluation and Governance 2025