The video discusses the foundational protocols enabling AI agents to operate at scale, highlighting six key protocols launched in the past year: MCP, A2A, AGUI, A2UI, AP2, and X42. These protocols form the substrate that shapes the customer experience in agentic systems. Among these, three—MCP, A2A, and AGUI—are emerging as the core agent stack, addressing critical questions about what tools agents can use, how agents coordinate with each other, and how humans maintain control over agent workflows. The other three protocols, while important, remain contested or domain-specific and are not yet part of the core standard.
MCP (Machine Common Protocol) is the most widely adopted and addresses the challenge of enabling agents to access and invoke external tools and data sources seamlessly. Before MCP, integrating tools like GitHub, Slack, or Salesforce with AI agents required custom, complex glue code. MCP standardizes this interaction, allowing agents to discover and use tools without rebuilding connectors for each platform. However, MCP assumes a high-trust environment and does not inherently solve security concerns, such as tool poisoning attacks, which require additional layers of security, approval flows, and audit mechanisms.
A2A (Agent-to-Agent) protocol tackles the problem of delegation and coordination among multiple agents, recognizing that no single agent can handle all tasks. It introduces the concept of an agent card, which describes an agent’s capabilities and interaction terms, enabling agents to discover and delegate work across product and company boundaries. While this enhances workflow flexibility, it also introduces complexity in terms of latency, failure handling, and permissions, making it suitable only for workflows that genuinely require distributed expertise or authority.
AGUI (Agent Graphical User Interface) focuses on the human control layer, enabling users to observe, approve, and interact with long-running, non-deterministic agent workflows. Unlike traditional web apps or chatbots, AGUI supports streaming state, event handling, and real-time approvals, addressing the supervision challenges posed by autonomous agents. This protocol is crucial for building trust and ensuring that humans can effectively steer agent actions, making it an essential part of the core agent stack alongside MCP and A2A.
The remaining protocols—A2UI, AP2, and X42—serve more specialized roles. A2UI provides a secure way for agents to render structured user interfaces without executing arbitrary code, while AP2 and X42 address different aspects of agent-led payments and commerce. AP2 focuses on user authorization and commercial trust in agent transactions, involving major payment players, whereas X42 enables machine-to-machine payments for resources without traditional account setups. These payment protocols are still evolving and reflect the complexity and importance of integrating secure, user-trusted financial interactions into agent workflows. Overall, the video emphasizes the need for builders to deeply understand these protocols and their implications on customer experience to successfully develop scalable, trustworthy AI agent systems.
