wdym by "beyond comparing description embeddings"? thats not how agents currently work. they currently work by asking the llm to pick a tool, very similar to this approach

→ View original post on X — @hwchase17