Third assembly line point · Technological HeterogeneityAgent to Agent (A2A)

The A2A protocol allows different AI agents (e.g. from ADK, CrewAI, LangGraph) to "speak the same language" (a common protocol) and follow specific processes to complete tasks together.

🧔End-User
💻Frontend &
Host Agent
Client/Router AgentA2A Client (Cameriere Agent)
A2A Protocol
A2A Servers (Remote Suggeritore Agent)

Specialized Agent 1 (e.g. ADK)

Specialized Agent 2 (e.g. LangGraph)

Specialized Agent N (e.g. CrewAI)

Agent Card

A machine-readable JSON metadata file that acts as a digital "business card" for an AI agent.

.well-known/agent.json
Agnostic

Sub-agents are tied to the specific AI library/framework language (e.g. Python wrappers) of the main agent, A2A is not.

Infrastructure Flexibility

Unlike agent/sub-agents, code updates happen strictly inside the agent that is involved, keeping the others untouched.

Independent Scalability

Agent and sub-agents share the same runtime/container, forcing monolithic scaling for heavy workloads. In A2A, agents scale independently.

Security

Agent and sub-agents share an execution environment. In A2A, agents are isolated from each other: if one is compromised or misconfigured, the others are not.