BAML: a new programming language for AI
summary
A new language claims to turn LLM calls into typed functions so models behave like normal code with BAML files, test cases, a run button and flexible parsing instead of brittle prompt hacks. Demos show Cursor integration, GPT-4o Mini selection, Python/TypeScript client code and cross-language type coherence. It runs offline, is open source, reports raw web requests, enforces type constraints instead of retries, and already lists over a thousand companies processing petabytes in production.
AI-written summary · about
chapters (13)
- 0:00AI models reliable
- 1:09Software should not fail
- 2:37React changed web dev
- 3:41Jupyter defines data science
- 4:44LLM calls as type systems
- 7:09Agents are while loops
- 8:52Components as composable functions
- 9:13Introducing BAML
- 13:47BAML integrates with languages
- 15:31AI not restricted to Python
- 16:13Reliable AI pipelines with BAML
- 17:17Why BAML is easy
- 18:19What BAML stands for
related talks
36mBAML - A new programming language for building Agents
15mBAML - A New Programming Language for Building Agents
27mAI Demystified Unveiling the Intelligence Behind the Machine
34mProgramming Languages for AI Programing Agents
57mArtificial Languages are Dead. Long Live Artificial Languages!
47mBuilding Python based AI Systems with LLMs
26mLanguage Model Agents Enable Semi-Formal Programming
43mFrom Lambda to LLMs: AI Coding and the Future of Programming