Notebooks, training jobs, and eval harnesses are Python even when the product is TypeScript. Packaging, GIL limits, and dependency hell are the tax. Teams still pick it because the ecosystem is where the papers' code actually runs.
The dominant language for AI/ML because of PyTorch, NumPy, and Hugging Face — with hot paths often in C++/CUDA/Rust.
Notebooks, training jobs, and eval harnesses are Python even when the product is TypeScript. Packaging, GIL limits, and dependency hell are the tax. Teams still pick it because the ecosystem is where the papers' code actually runs.