Indeed, which comes with its own advantages and disadvantages. As someone that has to always care about performance, Machine Learning, I am very happy not to have to worry about dependencies not being compiled to take advantage of the latest and greatest CPU instructions on a specific box in a very heterogeneous cluster. Now, you could just compile it all on each deployment, but something as large as say TensorFlow takes ages and not everyone gets to enjoy having great devops at your beck and call.