Great article - and I strongly agree. Notebooks are fantastic for experimenting, but once a final solution is found, it should be turned into standalone, modular components before making it into production.<p>I think the tough part is that most modern data workflow tools don't work well for data scientists. They can't quickly deploy and experiment with code variations on the cloud, so they end up in continuous experimentation mode with notebooks where they can make line-by-line tweaks. If the tooling could facilitate quicker deployment, it might be easier to break the notebook habit.