MCP solved tool calling. A2A solved coordination. What solves transport?

The article argues that AI agent infrastructure is in a familiar early phase where many protocols emerge before the market consolidates around interoperable standards. It says protocols such as MCP, ACP, A2A, and ANP are often framed as competitors, but may actually address different parts of the agent communication stack.

The key architectural question is less which protocol wins outright and more how the ecosystem defines clean boundaries between tool use, coordination, identity, and transport. If standardization happens too slowly, developers may face fragmented agent networks; if it happens well, agent interoperability could follow the same consolidation path as REST, WebSockets, and MQTT.