"We must make Photoshop dramatically more configurable," Nack said. "Presenting the same user experience to a photographer as we do to a radiologist, as to a Web designer, as to a prepress guy, is kind of absurd...With the power of customizability, we can present solutions via task-oriented workspaces"
"We're already making the code modular so that people aren't running what they don't need," Nack said. "Now we need to follow up at the user experience level, so that people don't have to wade through anything not geared towards the task at hand."
interesting, because that's one of the goals with plasma for the desktop. some work was done last night to get the zooming working a bit more the way we want it to, and that'll be an integral part of this approach. though that is probably not directly applicable to photoshop type apps, the idea of zoom-n-pan for tool discovery might not be a completely bad idea to experiment with.
Nack also noted that this won't result in a faster application when measured in clock-cycles, but that the improvements will come when measuring user interaction and perception. when one has to choose between the two, making the user move faster is more important than making the software move faster. that is because the point of software used by people (versus software that cranks through computations or transactions) is to benefit the person using it. making the computer work harder so the user doesn't have to is really the entire point. i feel we kind of lost track of that idea over the last 20 or so years of client computing.
to head off the obvious comments here: this isn't an excuse to ignore performance, but it is a guideline for when to sacrifice performance. at the same time, there are many techniques that can be applied to software designs of this sort that keep them reasonably performant, especially when the design allows software to be easily targeted towards specific use cases (meaning less baggage carried along the way and distancing expensive assumptions away from the core).