Build an A2A Agent

• ✓Explain why two independently-owned agents need a protocol instead of a shared function call
• ✓Write an A2A agent card advertising what an agent can do and how to reach it
• ✓Build an agent that wraps an existing MCP server as its own tool backend
• ✓Implement and track the A2A task lifecycle, including polling for a non-instant result
• ✓Have one agent delegate a sub-task to another agent mid-task and merge the result
• ✓Distinguish a deliberate, inspectable task failure from an unhandled-exception crash and from the dependency simply being unreachable, and handle each honestly
• ✓Score a multi-agent system against a seeded set of known-correct answers, not a single demo

Let two agents owned by different teams, services, or companies interoperate on a real task without sharing code or trusting each other's internals — the actual integration problem the moment an agent system grows past one process.