Not really, and this is what I meant by having an order of magnitude improvement is very exciting to the researcher even if in practice there are still 29 to go, but it doesn't make a big impact. Even something as simple as the borrow checker -- that only brings you to about where most of software, with all its bugs, already is -- comes at the cost of only being able to use memory in certain ways. I.e. to make the property inductive, it had to be made more coarse (indeed, that is what is always done when looking for inductive properties [1]) which requires more effort on behalf of the user.

As I've written above, over the past two or three decades we've only learned the theory that things are more difficult than we thought, and at the same time the capabilities of sound verification have fallen further away from the needs of mainstream software. The trajectory of both the theory and practice of sound methods has been negative. All the while, we've made some nice strides in unsound methods. Tony Hoare started recognising it in the nineties (which somehow shifted his 70s view on the subject), and the trend has continued: Yes, there are improvements in sound methods, but the problems become harder faster than those improvements, and the improvements in unsound methods is also faster.

[1]: That is seen very elegantly in TLA+, but I digress.