- Can we determine if/when a multi-agent system will loop around (other than through an observability platform), and can a supervisor node reject output until it has gone through a specified number of loops?
- Is there a way to measure if and how much better or worse the output is given different recursion levels?
Yes, it is possible to determine if and when a multi-agent system in LangGraph will loop around without relying solely on an observability platform (e.g., LangChain tracing via LANGCHAIN_TRACING_V2
). This can be achieved through the design of the graph’s structure, state management, and conditional edge logic, as demonstrated in the notebook and supported by the class materials.