Programming may someday escape the need for precision. The "compilers" of the day may be able to offer meaningful defaults and fitting. This sounds like full-blown AI, but it doesn't need to be, just as Google looks like it but is not. In fact the way to do it may be some kind of statistical approach relying on tones of code already written.
It's the need for precision that brings out the complexity. In my experience working with domain experts, they usually don't realize how complex their domain is. Only when we start trying to nail things down precisely do the contradictions and edge cases come out of the woodwork. Invariably it's order(s) of magnitude harder than it looked at first. Usually they are surprised and say they had no idea there was so much to do.
So while I agree that domain experts are accomplished at handling complexity, they do so the way that humans do it (ambiguously and inconsistently), not the way that computers do it (through formal specification). The ultimate bottleneck in software development is that there just aren't that many humans who can do the latter well.
I agree that if programming ever does escape the need for precision, then the game changes, completely and unrecognizably. A pet belief is that this would require computers that work more like the way humans do (down to the lowest level), but that's a wild guess. In any case, I haven't seen any evidence of things progressing in this direction, have you?
For some dramatic evidence that imprecise is better and it really works, look at the success of Google and the non-delivery of the Semantic Web. Google's approach is probably best explained by Peter Norvig in his "Theorizing from Data" lecture:
http://www.youtube.com/watch?v=nU8DcBF-qo4
State of the art in natural language processing is achieved with statistical inference, not building precise linguistic models. See e.g. Marti Hearst's text here:
http://www.edge.org/q2008/q08_7.html
Back to programming languages, I believe that a lot of the Ruby/Rails success is due to mimicking natural language in DSLs. Rails goes as far as using pluralization. DSLs bring convenience at the price of less understanding: people tend to try using them intuitively, without understanding precisely what happens, much more so than they do with libraries/frameworks.
Perl, similarly, was created with linguistic principles in mind:
http://www.wall.org/~larry/natural.html
It is by no means conclusive evidence, but I believe there is an underlying trend.
The program was based on the notion of mapping flows as topological bundles, thus it could (and would) attempt to automatically parallelize whatever independent operations it could. Programs could be set up as, quite literally, flowcharts, with a given box having input hooks (multiple), output hooks (usually single), and a piece of code, usually a C++ library, that would produce or consume the data flowing through the 'pipes'.
I demonstrated this to a bunch of high school kids at a national science fair (ran it on an SGI in Ithaca while presenting at the fair in DC... it was hard to compete with the UIUC Cave VR developers otherwise) and the kids seemed to understand it instantly. Hey, look at that, functional programming in a visual style. What's not to like.
Needless to say it has not set the scientific programming world on fire and the project, while still maintained and used (I believe) by a handful of groups, isn't terribly high profile. A screenshot can be seen on OpenDX.org : http://opendx.org/images/opendx-screenshot.jpg
I 'contributed' the RPM spec file, played with it for a while, but by the time it had been opensourced, I was no longer working in the type of environment where it was a major advantage. Plus I no longer had a pile of nodes to throw at arbitrary computational task nor much need.
Sometimes I miss that sort of thing. Lots of times, actually. I've been daydreaming about it lately... although these days I like to use iPython for that sort of thing, it would be nice if the self-documenting aspect of the UI and the generated programs could be retained.
Anyways.
I think these sorts of "boxes and arrows" programming systems suffer from being monolithic environments with idiosyncratic interfaces and little or no support for external software engineering tools. What's worse for them is they generally compete for mindshare with scripting languages or scriptable environments (i.e. things like Mathematica) with larger communities.
I don't think they will ever gain wide acceptance until they allow the same level of reuse as systems based on structured text.
I'm not saying that Simonyi won't come out with anything of value. Maybe it will help to facilitate dialogue between programmers and domain experts about how a system should work. But the hype that it will turn non-programmers into programmers is pure PR.