I honestly feel like this is a better idea in general, at least in data science stuff. For context, I tend to get dumped with undocumented poorly written untes

I honestly feel like this is a better idea in general, at least in data science stuff. For context, I tend to get dumped with undocumented poorly written untested DS code bases every time I move to a new job. The difference between setups where there's a bare VM/server and it runs using standard tools (command line scripts) or at least some open-source orchestration and any kind of proprietary technology tied to infrastructure is massive. Additionally, the server stuff can be maintained and debugged by a far larger number of people, thus reducing bus factor. Like, maybe in the web dev world there are things that everyone knows that are AWS services (maybe S3 counts, I guess) but in the DS world, I just know that I'm in for a whole lot more pain if they haven't just used bare servers.