Hierarchical vs. Plan-and-Execute · AI Developer Accelerator

May '24 (edited) • CrewAI

Hierarchical vs. Plan-and-Execute

I'm wondering whether anyone's read the Plan-and-Execute Agents blog post or watched Adam Lucek's Breaking Down & Testing FIVE LLM Agent Architectures - (Reflexion, LATs, P&E, ReWOO, LLMCompiler) video based on it. I'm wondering if these plan-and-execute processes have been attempted in a crewAI system using LangChain calls instead of using the crewAI hierarchical manager llm. I've tried hierarchical crewAI without gpt-4 as the model and have only gotten errors (unquantized llama-2-13b). I'm wondering if defining these processes in LangChain would provide a better output for those of us who aren't allowed to use external inferencing services.

Anyone already experimented with this. or would I be blazing a new trail?

1 comment