Microsoft’s Autogen vs CrewAI - tested on a diverse range of use cases

---

---

AI Summary

Summary: Multi-Agent Frameworks Comparison

Introduction

• Hasagi compares two leading multi-agent frameworks: Autogen (Microsoft Research) and CRE AI.
• Evaluation based on parameters like ease of use, quality, speed, scalability, customization, and code execution.

Getting Started

• Autogen: Simple setup, clear instructions, and code runs on first attempt.
• CRE AI: Getting started guide less straightforward, too many dependencies, and complex initial example.

Internet Search Capability

• Autogen: Easy integration of search APIs, good documentation.
• CRE AI: Built on L-chain, intuitive design, but relies on external tools for defining capabilities.

Writing a Blog Post with Multiple Agents

• Autogen: Seamless process, conversational format.
• CRE AI: More structured and templatized, easier for beginners.

Code Execution

• Autogen: Can execute code, handle errors, and generate outputs like plots.
• CRE AI: Can write code but cannot execute it.

Scalability

• Autogen: Handles concurrent requests well, caching improves performance.
• CRE AI: Slower, does not scale as well, needs improvement.

Final Report Card

• Ease of Setup: Autogen 4, CRE AI 3 (Functional but inconsistent documentation)
• Tool Calling: Both perform decently
• Human in the Loop: Both allow human input
• Customizations: Autogen 4 (more flexible), CRE AI 3 (relies on L-chain)
• Code Execution: Autogen 5 (supports execution), CRE AI 0 (cannot execute code)
• Scalability: Autogen 4, CRE AI 3 (needs improvement)
• Speed: Autogen 4 (faster), CRE AI 3 (slower with more dependencies)
• Documentation: Autogen better than CRE AI
• Quality of Output: Both similar
• Learning Curve: CRE AI easier, Autogen requires understanding of terminologies

Conclusion

• Autogen: Preferred for complex use cases with code execution.
• CRE AI: Suitable for simpler projects and quick starts.

Recommendations

• Share perspectives on the frameworks.
• Subscribe and like the video for more content.
• Comments are welcome for further discussion.