The idea is:
Re-integrate the ReAct Agent and Plan and Execute Agent as dedicated nodes, similar to the existing Information Extractor or Text Classifier nodes. These nodes would serve specific use cases like multi-step reasoning, planning, and dynamic tool usage, making them easier to implement without manual LangChain code.
My use case:
I’ve been developing a self-assisting AI agent using cascaded agents such as Plan and Execute and ReAct. These agents were critical in building autonomous behaviors, where the AI could break down tasks, reason through steps, and call tools dynamically to achieve complex goals. After the deprecation of the “Agent Type” field, replicating this logic now requires much more effort using lower-level components.
I think it would be beneficial to add this because:
- Reintroducing these as dedicated nodes would strike a balance—keeping things simple for most users, while still empowering advanced users to build intelligent agents with minimal effort.
Any resources to support this?
- Deprecation of Agent Type in AI Agent Node – GitHub PR #13459
- Examples of how Information Extractor and Text Classifier nodes make advanced tasks user-friendly.
Are you willing to work on this?
I’m happy to share more about my use case, test new features, or contribute ideas to help shape the development of these nodes.