Bob Sutor's opinion piece for Loosely Coupled this week, Demystifying ESB, is based on a recent posting to his weblog. Perhaps the most surprising thing about this article is the discovery that Bob keeps a blog, which inadvertently seems to have been something of a well-kept secret; when I subscribed yesterday on Bloglines, it only had one subscriber.
The essence of Bob's argument in his opinion piece is that ESBs are built, not bought and that therefore, any organization with an existing messaging infrastructure already has an ESB, or at least the makings of one. I think this is a valid argument, and it has an interesting corollary, which is that buying an ESB doesn't obviate the need to build one. Or at least, just because you buy an ESB, don't imagine you won't have to tack other stuff onto it (potentially even other packaged ESBs) before your service bus/fabric infrastructure is complete.
The same, of course, applies to SOAs. It may sound attractive to buy an SOA, ready-made and prepackaged, from your favorite (or incumbent) software platform vendor IBM, for example, or BEA, Oracle, SAP, whoever. But that won't obviate the need to get it working with other parts of your infrastructure (potentially even other packaged SOAs) before your services architecture is finished.
I'm sure that, maybe a decade or more from now, it will ultimately become possible to buy plug-in SOA components off-the-shelf. But that will only happen once all the standards are baked and everyone has ironed out the wrinkles in all the different implementations. In the meantime, to use Bob's house-wiring analogy, everyone is going to have to resign themselves either to mugging up on all the circuitry and componentry or else hiring professional fitters who know all the ins and outs of connecting up disparate pieces of services infrastructure. This is emerging technology, and there are no short cuts.