At first glance, Pipes looks really cool. To me, it gives an excellent graphical / symbolical view of what it is all about: connecting data processors, input to output. Indeed, much like Unix pipes.
That said, it also reminds me an awful lot of an old “graphical database application” (forgot the name), that I was forced to use on the Mac, way back early 1990s. The idea was that people without any programming (and database) knowledge could created and drag/drop their own database application.
There was a serious problem, however: the tool didn’t work at all. Programmers missed even the most elementary tools (no loops, no exit conditions), while people not into programming are not inclined to develop an abstract view of their data. Which was exactly the reason I got the assignment.
The flaw is that this class of products are aimed at a non-existing audience. At one hand, we have complex tasks, which can be solved by expert users using expert tools. At the other hand we have lay users, wanting to get the results of complex operations – which is perfectly reasonable. However, requirements for these kind of graphic tools are extremely tough. As an example, providing a graphic wrapper around an expression which is basically a regexp will not help anyone.
Bottom line: I suspect that those people who instantly grasp the meaning of the graphic enhanced tool would prefer their own (command line) tools, while the for the rest of us it is still way too abstract and thus complex. Provide real abstraction hiding, or you still need power users which will ignore your enhanced tool anyway.
More recent examples of this flaw are in my opinion Apple Automator and the whole Applescript language. When it comes to exact, abstract object properties, I prefer to just say window.location.search, rather than “the search property of the location of the window” or something equally fuzzy. But I digress.
So, back to Pipes: yes, there’s much eye candy to its lovely interface, but when it comes to pure production value, I prefer more abstract tools which give me more control. Hide all geeky abstraction involved, and it might evolve into something really great – way to go.
For now, mixing and matching RSS feeds could be as simple as using command line xsltproc with a simple XSL tranformation.
(maybe there is a niche for it in the cut-n-paste crowd – those who just google some code, paste it somewhere and have it magically work, sometimes. Pipes will at least give them a robust and well defined environment, where they will not easily shoot themselves in the foot).