Their self-driving engineers are now split among three locations: San Francisco, Pittsburgh, and Arizona. I think that alone is going to cause significant engineering friction. It's a lot harder to do remote work for testing live robots (especially if they can kill people).
The Pittsburgh team is there because of Carnegie Mellon, so likely where the hardware gets made. The software engineers in SF just need data from vehicles, which doesn't require cars to be near them. The operations folks actually going out and testing the cars can be anywhere, as long as they can stream data back to the SW and HW folks.
The self-driving car ML software is also being worked on in Pittsburgh, along with the hardware.
Source: I grew up in Pittsburgh, and now have lived in SF since graduation.