Microsoft Alumni Reflect on Building AI Agent Frameworks in Open Source
Key Takeaways
- ▸Early AI agent frameworks were brittle, but improvements in models, tooling, and abstractions are making systems more robust in 2026
- ▸Microsoft's open source projects Semantic Kernel, Autogen, and GraphRAG represent significant contributions to AI agent architecture
- ▸Building AI systems has become dramatically faster than even a few years ago, though challenges around memory and orchestration remain
Summary
In a podcast episode of 'Humans of AI,' host Alex Chao interviews Eric Zhu about their experiences building foundational AI agent frameworks at Microsoft, including Semantic Kernel, Autogen, and GraphRAG. The conversation explores the technical evolution of AI systems, addressing why early agent frameworks proved brittle and how improvements in models, tooling, and abstractions are changing the landscape in 2026.
The discussion covers critical architectural decisions around memory management and orchestration in AI systems, while acknowledging that building AI products has become dramatically faster compared to just a few years ago. Both builders share insights from hands-on development experience in the open source space, offering perspective on what frameworks and architectures are powering modern AI products today.
Beyond technical details, Chao and Zhu address the human dimension of working in AI, discussing career navigation strategies for AI engineers and researchers, avoiding burnout, and managing the challenges of working in such a rapidly evolving field. The conversation aims to explain the 'why' behind modern AI system design choices, making it relevant for AI engineers, founders, researchers, and those curious about the direction of AI development.
- Career longevity in AI requires strategies for managing burnout and staying current in a rapidly evolving field
