I watched this critical video of the CrewAI agent framework (as well as other frameworks like AutoGen) and I’m now rethinking a few design ideas on how to use agents.
The main issues are that these frameworks can be slow, expensive, and inconsistent. These issues will eventually be resolved with more powerful and cheaper models, but for now, they are weaknesses that limit how much these tools can be used in real high-stake use-cases.
The author of the video has a strong opinion on what comes next for agentic frameworks: He doesn’t believe these frameworks have a future! Why? It’s very likely that more powerful LLMs are going to be able to complete complex tasks directly, without the need to run complex tasks through layers of multiple agents.
Here’s the video.