What's this about?
Modular open-sourced Mojo, its programming language for AI systems work, under Apache 2.0. HN immediately supplied the nuance: it's source-available now, with a commitment to accept outside contributions early next year — though the Apache 2 licence already permits forking today.
What exactly is Mojo for?
The AI world has a two-language problem. Python is what everyone writes because it's pleasant, but it's far too slow for the numerical work underneath, so the actual computation lives in C++ and CUDA. That means the people designing models and the people making them fast are working in different languages, and crossing that boundary is where much of the pain lives. Mojo aims to be one language for both: Python-like to write, but compiling to code fast enough for the kernels — with an ownership system adjacent to Rust's, compile-time evaluation similar to Zig's, and a first-class dependent type system, built on LLVM.
What's the community saying?
- The licensing distinction was the first correction, and made carefully rather than cynically — source-available with a promised contribution path is not nothing, and Apache 2 already permits forks.
- "As long as it was proprietary, I could ignore it." The comment that captures the actual change: openness moved Mojo from a curiosity to something people feel obliged to evaluate.
- A genuinely interesting argument for this licensing shape: a permissive licence without open upstream contributions is SQLite's model, and one commenter suggested it will spread partly because it shields projects from a flood of AI-generated contributions.
- The practical questions were all about the ecosystem — getting-started resources, and whether there's anything NumPy or SciPy-like yet, which is the real adoption gate.