
Languages designed for LLM coders?

From Swift to Mojo and high-performance AI Engineering with Chris Lattner
Topic 13 of 14
Published Nov 5, 2025 · Clip 1:25:13–1:28:58 (3:45)
Chris argues optimizing languages for LLMs is the wrong goal because code is read more than written and AI has made writing easy; what matters is the intersection of expressivity (can you express the hardware's full power?) and readability. Mojo embraces Python's readable syntax while replacing the implementation to express full hardware power. On agent-facing concerns like error messages, he says what's better for an agent is also better for a human, and notes the massive open-source Mojo codebase (~700k lines with full history) is what really helps LLM-based coding.