The AI agent ecosystem has exploded with options—Google's Agent Developing Kit, Claude SDK, OpenAI Agents, LangGraph, Pydantic AI, and more emerging monthly. But which one should you actually use? This hands-on workshop cuts through the hype to give you a practical decision framework for selecting the right agent SDK for your use case.
We'll build the same production-grade agent using three different approaches, revealing the real tradeoffs in developer experience, flexibility, and operational complexity. You'll learn when graph-based orchestration makes sense versus simpler frameworks, how vendor SDKs compare to framework-agnostic solutions, and what questions to ask before committing your team to a specific path.
Through live coding and architecture discussions, we'll examine critical factors practitioners face: integration with existing stacks, debugging and observability, cost implications, team learning curves, and avoiding vendor lock-in. You'll walk away with a reusable decision matrix and hands-on experience that translates directly to your next agent project.
Key Takeaways
1 Decision framework for evaluating agent SDKs based on your constraints
2 Hands-on comparison of 3 major approaches through live code
3 Production considerations: monitoring, cost management, and team scalability
4 When to use graph-based vs. linear orchestration patterns
5 Strategies for future-proofing your agent architecture
Speaker
Hien Luu
Sr. Engineering Manager @Zoox & Author of MLOps with Ray, Speaker and Conference Committee Chair
Hien Luu is a Sr. Engineering Manager at Zoox, leading the Machine Learning Platform team. He is particularly passionate about building scalable AI/ML infrastructure to power real-world applications. He is the author of MLOps with Ray and the Beginning Apache Spark 3 book. He has given presentations at various conferences such as MLOps World, QCon (SF,NY, London), GHC 2022, Data+AI Summit, XAI 21 Summit, YOW Data!, appy()
