Steve's post on "the modeling problem" hits the nail on the head. We're all familiar with the concept of "fast, good, cheap - pick two". Steve breaks down modeling into "general, precise, efficient - pick two (and favor one)". Furthermore, you can't have a language that is both general and precise. UML takes what Steve calls the "Favor efficiency, accept generality and compromise precision" approach:
The UML metamodel is flexible enough to allow it to describe virtually any system out there. However, from a formal semantic perspective, the resultant model is gooey and formless which makes it very difficult to compile into anything useful. At best, we can get some approximation of the underlying system via codegen, but even the best UML tools only generate a fraction of the code required to fully realize the model. The lack of precision within the model itself requires operating in both the model domain and the system domain, and implies that some facility exist to synchronize the two. Thus, the imprecision of UML forces us to solve the round-tripping/decompilation problem with 100% fidelity, which is generally difficult to do.
Software Factories, on the other hand, takes what he calls the "Favor efficiency, accept precision, and compromise generality" approach:
This, I think, it the sweet spot for Microsoft’s vision of Software Factories. Here’s why: the classic problem faced by modeling languages is Turing equivalency. How do you model a language that is Turing-complete in one that’s not without sacrificing something? The answer is: you don’t. You can either make the modeling language itself Turing-complete (which sacrifices efficiency) or you can limit the scope of the problem by confining yourself to modeling only a specific subset of the things that be expressed in the underlying system domain. Within that subset, it might be possible to model things extremely precisely, but that precision can only be gained by first throwing out the idea that you’re going to be able to efficiently and precisely model everything.
When describing Software Factories, I have two analogies that I use to explain the idea. The first is the "houses in my neighborhood" example I blogged before. That does a good job describing economies of scope, but doesn't really cover the modeling aspect of software factories. Talking about how you model cars or skyscrapers doesn't really capture the essence of software modeling - you don't generate the construction plans from a scale model of a skyscraper. However, it turns out that all developers have at least a passing familiarity with my second analogy: Visual Basic, the most popular DSL and modeling tool of all time (so far).
The original Visual Basic was a rudimentary software factory for building "form-based windows apps". (Today, VB.net has been generalized to support more problem domains) Like the factory approach that Steve describes, VB was very efficient, sufficiently precise, yet not particularly general (especially in the early years). There were entire domains of problems that you couldn't build VB apps to solve. Yet, within those targets problem domains, VB was massively productive, because it provided both a domain specific language (DSL) as well as a modeling environment for that domain.
A DSL incorporates higher-order abstractions from a specific problem domain. In the case of VB, abstractions such as Form, Control and Event were incorporated directly into the language. This allowed developer to directly manipulate the relevant abstractions of the problem domain. Abstractions extraneous to the problem domain, such as pointers and objects in this case, got excluded, simplifying the language immensely. Both of these lead directly to productivity improvements while limiting the scope of the DSL to a particular problem domain.
In his post, Steve makes the point that it's pointless to distinguish between modeling and programming languages. VB certainly blurred that line to the point of indistinguishably. Regardless, graphical languages are typically more compelling and productive than textual ones. It's hard to argue with the productivity that VB form designer brought to the industry. Dragging and dropping controls to position them, double clicking on them to associate event handlers, changing properties in drop down boxes - these idioms have been widely implemented to the point that essentially all UI platforms provide a drag-and-drop based modeler. It's such a great design that 10 years later, UI modelers are essentially unchanged.
Once you realize that VB's DSL and modeling environment was a rudimentary software factory, you realize that Software Factories methodology is about generalizing what VB accomplished - building tools that achieve large gains in efficiency by limiting generality. Since each of these tools focuses on a limited problem domain, you need different tools for different problem domains. The problem is that while building apps with VB may be easy, but building VB itself was not. Most enterprises have the expertise to develop abstractions in their domain of expertise and to codify those abstractions in frameworks, but very few can develop tools and DSLs for manipulating those frameworks. One of the goals of Software Factories (and VSTS Architect for that matter) is to make it easier to build tools that are really good at building a narrow range of applications.
Note, it's important to note that the term "narrow range" is relative. Darrell seems to think narrow range only means vertical market applications that don't "solve new and interesting problems". It's true that the narrower the range, the more productive the tool can be. But VB shows us that you can achieve large productivity gains while solving new and interesting problems even in broad scope problem domains.