Yeah I mostly agree with this. There are some small wins though like how BigML allows you to export random forests into plain text node.js and you can just deploy in shitty php pipelines and boom you're doing machine learning.