Each small bump in model performance kills another agent framework. The base LLMs can do so much in-context and if they need more internal structure, they can write tools. You don't need the framework. The only thing frameworks should be doing is making the problem easier for the LLM to solve.
when I left CrewAI at the end of last year, a lot of friends asked why.
it didn't make sense to most. agent frameworks were hot. funding was flowing. why leave now?
the main reason: i saw this squeeze coming. some might disagree, but this was and still is my thesis.
Each small bump in model performance kills another agent framework. The base LLMs can do so much in-context and if they need more internal structure, they can write tools. You don't need the framework. The only thing frameworks should be doing is making the problem easier for the LLM to solve.
hard agree on this.