There's a difference between can't and won't. Yes their operation in Phoenix is heavily mapped and fenced, that doesn't mean that their technology can't and hasn't driven in more improvised environment before, and I highly doubt the lumbar street example was mapped.
Videos recorded by the company will always be cherry picked by definition, since they would only post videos that show the technology in its optimal working state. And there's always a driver behind the wheel checking everything goes right.
The reason Waymo is operating the way it does is they realize that they need to be 120% cautious, as ONE accident is all it takes to throw all their work a decade behind, jst like one accident is all it took to completely kill Uber's self driving effort. So again, it's not that they need a fully mapped geofenced area, it's that they choose to use that for now.
Hell, Tesla themselves posted their first "A to B" video back in 2016, yet here we are 5 years later and from a high level, what they showed in that cherry picked video they posted isn't much different to an average viewer, but no one would claim that they haven't improved at all in the past 5 years.
This is basic anecdotal data vs real statistics. A video tells you nothing about where the technology is. Unless we can see real open data about miles driven and number of disengagements, there's no way to tell if Tesla is doing well or not. So far Waymo's numbers are an order of magnitude better than any competition.