I Tested AI Agents Team Using the CrewAI Framework
I finally got around to trying out the CrewAI framework hands-on, which is designed for team-based agent interactions to solve use cases.
My goal was to personally learn how to use technologies for creating AI agents for use cases.
As a use case and team architecture, I simply took an example from CrewAI’s GitHub. Within this example:
- Two agents: one researcher, the other a writer
- Tasks: 1) conduct market research on the latest AI advancements, 2) write a summary
The agents interacted and delivered a result (screenshot attached). The quality of the result should not be judged, as this was a training iteration and no tools were used to enhance the quality, and the process itself was very basic.
Here are the technical skills needed to get started:
- Python — basic knowledge
- GitHub — basic knowledge
- VS Code — basic knowledge
I managed to do it, and so can you.
GPT-4 was used as the LLM.
The team of two agents for simple tasks turned out to be quite resource-hungry, and API requests ate up 18 cents from my OpenAI balance. You can’t just casually run it back and forth; you have to spend money very wisely.