Jason Liu. AI engineer and creator of Instructor, an open-source library that makes structured outputs from language models reliable enough to use in production systems. He writes at jxnl.co about applied AI engineering, agent architecture, and the practical problems that surface when deploying LLMs at scale. His work focuses on making AI systems that produce verifiable, typed outputs rather than freeform text, a constraint that separates production-grade agent work from prototype behavior. He also writes on the aesthetics and economics of creative work in the AI era, arguing that execution ability is now cheap and discriminating judgment about what is worth building has become the scarce input.
Operating themes
- Structured outputs as the reliability floor Typed, validated outputs from LLMs are the baseline for production reliability. Unvalidated freeform text is a prototype, not a system.
- Taste as the irreducible differentiator In a world where execution ability is cheap, discriminating judgment about quality is the scarce input. Developing it requires the friction that AI tools tend to remove.