They always try to put too much magic into the box, not realizing that all agents are, at their core, is just advanced text processing engines... data in -> process -> data out. Nothing more needs to happen!

What I did, was streamline this so that everything can be chained together without hassle

The big one is bad abstractions that do more harm than good, plus, way too many features which cause it to kind of collapse in on itself in terms of maintainability.

My framework rips out all external tools and makes them downloadable through a CLI instead + everything has input & output schemas

github.com/BrainBlend-A...

Atomic Agents is much less of a hassle... More ownership over your code due to the shadcn-like approach to tools, atomicity, and heavy use of the IPO model and streamlining all data input & output using #Pydantic

#Python #ArtificialIntelligence