You can already ‘pip install’ quite a bit (gensim being a prime example), but it takes some knowledge of how/why the general resource was trained to apply it to a business problem.

I think the big leap will happen when a "regular" developer (as in doesn't need to have any specific know-how), when they can just call simple APIs that hide all the complexity, Same way we don't need to know how a GPU works to render a red square in a browser window.

grandparent comment is saying that you _have_ to be able to apply the data and training techniques to a use case. what you're describing is the nocode fantasy where non-developers will be able to name complex apps, but applied to ml