Has CGI developed to the point that lots of stunt work can simply be done on computer?
Well, contrary to what many people assume, it's by no means "simple" to replicate real people or objects in a computer. It can only be done convincingly if you can devote a lot of time, effort, and processing power to it, which is why CGI effects can look all but completely convincing in big-budget motion pictures but still look cheesy on TV shows.
A computer is a powerful tool, but it makes no sense to depend on it exclusively rather than using it as just one tool in the box. The best way to create convincing screen illusions is to use a mix of different techniques, so as to keep the audience guessing, and simply to use whatever method works best for the needs of each particular shot.
So yes, CGI is to the point where it can be used for stunts that are too dangerous or difficult to achieve with live performers. We've seen it used for things like Spider-Man's swinging in the Raimi movies, Superman's flight in
Superman Returns, and big fight scenes in plenty of movies, and
sometimes it actually looks convincing. But if a stunt is something that a live performer can do reasonably safely, then it's usually going to be simpler and cheaper to do it that way, and more satisfying to the audience because they're actually watching something real and tangible.
One of the main ways CGI is used to enhance stunts these days is not to replace the stunt performers, but to enable supporting them with heavier safety wires and harnesses and such than in the past -- because these days the wires can just be digitally erased from the shot, and thus don't need to be invisibly thin. Also CGI can be used to replace a stunt performer's face with the face of the actor being doubled (something that was first done, as far as I know, with Ariana Richards's stunt double in
Jurassic Park), but it's still a live human being performing the stunt.