Today's guest is Jason Liu, creator of Instructor (https://github.com/jxnl/instructor). You might have seen him on Twitter (  / jxnlco  ) or at our AI Engineer Summit last year!

Instructor was downloaded over 150,000 times last month and helps thousands of developers create structured outputs from LLMs. It's used for data extraction, knowledge graph generation, as well as AI agents planning. We dove into all of those use cases in today's episode!

Show notes: https://www.latent.space/p/instructor

Timestamps:
00:00:00 Introductions
00:02:50 Early experiments with Generative AI at StitchFix
00:09:39 Design philosophy behind the Instructor library
00:13:17 JSON Mode vs Function Calling
00:14:43 Single vs parallel function calling
00:16:28 How many functions is too many?
00:20:40 How to evaluate function calling
00:24:01 What is Instructor good for?
00:26:41 The Evolution from Looping to Workflow in AI Engineering
00:31:58 State of the AI Engineering Stack
00:33:40 Why Instructor isn't VC backed
00:37:08 Advice on Pursuing Open Source Projects and Consulting
00:42:59 The Concept of High Agency and Its Importance
00:51:06 Prompts as Code and the Structure of AI Inputs and Outputs
00:53:06 The Emergence of AI Engineering as a Distinct Field