Short answer is build in HTML/JS/CSS/AJAX now. Feature detect and add HTML5 where you can. Polyfill with Flash where necessary. Think about the future.
It's true that HTML5 can't do everything that Flex can today. More important is the fact that the vast majority of enterprise apps don't require those things.
Developing an app in Flex just because it can do things today that HTML5 can't is short sighted. That rationale will not hold water 3 years from now when the boss is asking "what the hell is Flex?"
Working with Flex and AS3 is really nice on a lot of levels. The type-safety is a cozy blanket. The tooling is also excellent, and I love having access to integrated test runners and other handy bits that IntelliJ provides me.
If you consider Flex/AS3 as a DSL for building UIs, the story starts to get more interesting.
it quite different from my experience, I was using myeclipse + flex plugin. I remember that myeclipse hangs every 20 minutes, and that happen not only on my PC, that happen to all my teammates PC, while developing Flex, and it work OK when we're developing java backend.
I'm no fan of Eclipse. I've never used myeclipse, so I can't speak to that. I've used Flash Builder (in a mostly stock way) with varied levels of success. IntelliJ has really nice Flex tooling though.
It's true that HTML5 can't do everything that Flex can today. More important is the fact that the vast majority of enterprise apps don't require those things.
Developing an app in Flex just because it can do things today that HTML5 can't is short sighted. That rationale will not hold water 3 years from now when the boss is asking "what the hell is Flex?"