How Docker turned local AI inference into a first-class primitive, and why that changes how you build GenAI apps Three commands and an LLM is running on your laptop. No Python environment, no CUDA ...