And specifically, it's a lot harder than doing it in java or C++. I've seen C++ codebases that exceed 8 million lines of code and held together (not without friction, that I guarantee). I've seen java codebases with about half that that held together a lot better than the C++ ones.
Even accounting for the fact that python easily lets you achieve 5 times as much in the same amount of lines (at the cost, imho, of more surprises when wrong types get passed to your functions). I'm not sure "normal python" is more than a little bit more succint than (good) C++ though. Easier to read, sure, if you're allowed to make assumptions about types. But not that much shorter.
I've yet to see the first million line python codebase. And python has certainly had sufficient time to develop such codebases. Why aren't they here already ? Those C++ codebases were almost operating systems though (with extremely tricky threading support libraries, every project it's own of course), something you fundamentally can't do in python. The java codebases were financial stuff though, something that could easily be done in python, and they certainly have the money to absorb the increased hardware cost, and looking to have faster development turnaround. That's another discussion I can guarantee you will have with your boss if your app really grows, the hardware cost. When it comes right down to it, a programmer is worth (rule of thumb) about 15 servers. If python means one less programmer, but 50 extra servers, you will lose that discussion with your boss.
Dynamic typing is certainly a massive advantage for small codebases. It becomes more and more of a disaster on larger ones though. So an IDE for python ... it merely lets you lose the battle with your code that much faster.