I watched this a few weeks ago -- great talk and exciting tool!

Can you expand on this answer at the end about whether the generated code is run in production or not?

https://youtu.be/5YjsSDDWFDY?t=2200

I was surprised, because from a performance perspective, I'd expect that if you can generate code for USB drivers, you can also do so for cloud services? You have generally have better and more control over hardware in the latter case?

The state machines seem pretty low level, so I don't see in principle why efficient code couldn't be generated from them. Is it an issue of wanting the implementations to use many cores perhaps?

I wonder if the P-generated code is still used for USB drivers today, or if they moved away from that approach and just use the modeling process?

Thanks for any clarification

FWIW I'm using 3 or 4 C/C++ code generators for https://www.oilshell.org, and I've the layer of indirection useful in many ways (although it also can make it harder for people to understand)