Most of the time, it's "fast enough" for the job. But when it isn't, it really isn't.
Most of the time the standard libraries suffice for a job. But when they don't, you're suddenly in pip/easy_install hell. [1]
Most of the time you can get by with Python 3. But when you need to broadly distribute code with a dependency on a compiled binary, may the heavens help you. [2]
Most of the time, the syntax and "magic" won't bite you. But when it does, it does so hard. [3]
Most of the time, pylint/pychecker/etc. will find the same errors a compiler would...
You get the picture.
I still program with Python frequently, but I've increased the size of my programming language toolbox to compensate for its weaknesses.
[1] Or was that virtualenv, or setup.py, or apt/yum/nix, or...
[2] Broadly distribute means putting it on boxes where a C compiler is not available. This means you typically get to pre-compile those libraries and create either wheels or native packages. Not impossible, but for a "batteries included" high level language it's a lot of specialized low level grunt work.
[3] The infamous `def a(b=[]):` "feature" for one, anything involving metaclasses for another.