Use of Natural Language for Agent Communication

In current communication protocol designs, the prevalent practice is to define a strict Application Programming Interface (API) for different types of interactions among software entities. There is an explosion of APIs which makes interoperability hard, and the standards for interoperability fragile. Specifically for agent to agent communications, defining strict APIs for interactions such as registration, discovery, telemetry and debugging creates significant operational challenges. In this draft, we argue that a flexible design option -- where generative AI is used to create a flexible communication protocol for APIs, can lead to a common and flexible way for a single uniform API to be used between agents, and enables a better communication paradigm.