MicroAgents: Exploring Agentic Architecture with Microservices

The authors explore the concept of AI agents in the context of the Semantic Kernel orchestration. They define an AI agent as a modular abstraction that can possess a persona, perform actions in response to user input, and easily communicate with other agents. They propose the MicroAgent pattern, which utilizes microservices to partition agents by functional domain and enable agent composition.

To validate their approach, the authors conducted an experiment with eight critical APIs and twelve irrelevant APIs across five microagents. They found that both the monoagent and microagent approaches exhibited similar completion rates, but the microagent approach was particularly effective at coordinating complex function calling. However, both approaches sometimes had issues with tasks such as adding a trip to the calendar or booking a return trip.

The authors encourage feedback and further exploration of the MicroAgent pattern in AI development.