Why use Docker for Data Science Projects?

I am a beginner in Docker and just learning to integrate it into data science projects. From the information online, I see that Docker is advertised as a way to capture the libraries and dependencies of a project so as to solve the "it works in my machine problem".

What confuses me is that for most data science projects you are using Python + a limited number of libraries (tensorflow, sklearn, numpy,...), and so in practice you can already manage those using Python dependency managers like Poetry. These libraries don't require C++ compilers or fancy OS libraries, so why not just use Poetry? Docker seems like overkill.