The a16z Show How Foundation Models Evolved: A PhD Journey Through AI's Breakthrough Era
345 snips
Jan 16, 2026 
In this engaging discussion, Omar Khattab, an Assistant Professor at MIT and creator of DSPy, unveils how he unintentionally established a new framework for making large language models programmable. He critiques the conventional obsession with AGI and emphasizes the importance of building systems that bridge the gap between user intent and AI capabilities. Omar articulates the need for a middle layer between natural language and code, illustrating how DSPy captures intent through its innovative abstraction of signatures, enhancing the reliability of AI systems.
AI Snips
Chapters
Books
Transcript
Episode notes
Need A New Abstraction Between Code And Language
- Natural language prompts are too ambiguous and code is too rigid, so we need a middle abstraction to declare intent.
- Omar Khattab calls this shift from prompts to programmable intent like moving from assembly to C for AI systems.
PhD Timing Aligned With Foundation Models
- Omar described starting his PhD in 2019 just as foundation models began to emerge.
- He focused on how to actually use those models to build systems and applications during his PhD.
Systems, Not Just Bigger Models
- The frontier labs accept that scaling alone won't solve application needs and are building post-training pipelines.
- Omar argues we need systems (with retrieval, tool use, agent training) not just bigger models.
