I don't really find python slow for what I do (typically writing UIs around computer vision systems) but also, several years back I made a microcontroller-based self-balancing robot. It was hard to debug the PID and the sensor, so I replaced it with a Pi Zero and the main robot loop ran in python- enough to read the accelerometer, compute a PID update, and send motor instructions- 100 times a second. If there was a problem (say, another heavy process, like computer vision, running on the single CPU) it would eventually not respond fast enough and the robot would fall over.

Most of the time it's not that you need a faster language, it's that you need to write faster code. I was working on a problem recently where random.choices was slow but I realized that due to the structure of my problem I could convert it to numpy and get a 100X speedup.