Open Source Startup Podcast E183: Why English Isn't a Programming Language - the BAML Story
Oct 20, 2025
Vaibhav Gupta, Co-founder and CEO of Boundary, shares insights on creating BAML, a programming language designed for AI agents. He discusses the challenges of messy AI-generated code and how BAML offers a cleaner, more elegant alternative. Vaibhav highlights the importance of code quality and observability, emphasizing that beautiful code leads to fewer bugs and better functionality. The conversation touches on the limitations of using English for programming and why a dedicated language is essential for performance and debugging.
AI Snips
Chapters
Transcript
Episode notes
From Many Pivots To A Compiler
- Vaibhav and his co-founder pivoted through many ideas, including a Slack competitor and embedding optimization, before building a new language.
- They wrote a compiler in four days and saw a cleaner, more elegant way to express AI pipelines that motivated the pivot.
Pick Problems That Motivate You
- Optimize for problems that excite you day-to-day, not just short-term traction or revenue.
- Choose work where you get consistent, intrinsic motivation to sustain a long startup journey.
Stochastic Code Needs Stronger Tooling
- Stochastic LLM behavior makes code fragile and demands robust tooling like types and retries.
- BAML adds ergonomic features (dedenting, fewer escapes) that reduce errors and improve developer performance.