Speaker

Chris Lattner

Chris Lattner

American compiler engineer and programming language designer; creator of LLVM, Clang, MLIR, Swift, and Mojo; co-founder and CEO of Modular.

Lattner built LLVM as a university project in 2000 and joined Apple in 2005, where it became the foundation for Clang (Apple’s C/C++ compiler) and later Swift, the language Apple introduced in 2014. He spent time at Tesla, Google (where he created MLIR, the compiler infrastructure now embedded in virtually every major AI hardware stack), and SiFive before co-founding Modular in 2022. Modular’s thesis is that the AI hardware ecosystem — fragmented across Nvidia, AMD, Google, and a growing field of accelerator vendors — needs a portable, high-performance software platform that no individual hardware company is incentivised to build. Mojo is the language that makes that platform possible.

Core positions

The hardware-software boundary is where Lattner considers himself most useful, and his career follows that boundary upward as it moves. He holds that the right response to hardware fragmentation is not a ‘sufficiently smart compiler’ that hides complexity behind magic — such systems work on demos and fail under modification — but a language that gives programmers explicit control through a clean type system and a powerful metaprogramming facility. Metaprogramming (running code at compile time to generate specialised machine instructions) is the mechanism that lets Mojo deliver both portability across hardware vendors and full performance on each chip.

On language design, Lattner draws a sharp line between features that reduce the power-to-weight ratio and features that compound. He credits Go’s discipline of saying no as an underappreciated model, and cites Swift’s open proposal process as a cautionary tale: good ideas and feature bloat arrived together. He expects AI coding tools to dissolve the conventional ‘nobody adopts new languages’ assumption, because a large, public corpus of idiomatic code lets models assist in learning and migration without retraining.

In the wiki