Implement plan-and-execute” style agent to replace Reasoning and Action (ReAct)-style agents what we use now

[Abirdcfly]

ref: https://blog.langchain.dev/planning-agents/

[nkwangleiGIT]

it's to replace or keep both? not sure it'll do the same thing

[Abirdcfly]

Plan and execute agents promise faster, cheaper, and more performant task execution over previous agent designs.

It looks like the plan-and-execute approach is overall better than ReAct, and langchain themselves implement in https://github.com/langchain-ai/langchain/tree/master/libs/experimental/langchain_experimental/plan_and_execute, And they can be easily built by using LangGraph.

Since langchain is a library of tools, it provides all ways to do this, I guess we just choose the one that works best?

[bjwswang]

Plan and execute agents promise faster, cheaper, and more performant task execution over previous agent designs.

It looks like the plan-and-execute approach is overall better than ReAct, and langchain themselves implement in https://github.com/langchain-ai/langchain/tree/master/libs/experimental/langchain_experimental/plan_and_execute, And they can be easily built by using LangGraph.

Since langchain is a library of tools, it provides all ways to do this, I guess we just choose the one that works best?

It's a upgrade of current solutioni IMO. We can do some experiments.