To understand the future of software, we need to understand the past, because once again we're seeing that history is repeating itself. This article looks at how the next frontier of software ­ the dynamic assembly of Web services into a new type of composite Web application ­ is following closely in the footsteps of the Industrial Revolution.

The Artisan & Cottage Industries
The year is 1386. You're on pins and needles today. For almost three years you've been waiting for Merac, the village's master craftsman, to finish the ceremonial coat of arms he's been forging for your only son's wedding. As one of the most highly decorated knights in feudal France and a descendant of a bloodline dating back nine generations, you have the resources and wealth to pass on such a masterpiece to Philip. The only thing that can ruin this day is the weather near your castle outside Dordogne. It's raining. Again.

You finally lay your eyes on the handicraft ­ and your heart jumps. It is exactly as you imagined it; perhaps even more superb in person than what you described to Merac so long ago. You remember why you selected this artisan in the first place: he has painstakingly transformed your vision into a unique work of art. Philip will be delighted ­ and of course the townspeople will admire it for generations.

In 1386, stories like this abound. Craftsmen labor for their lifetime, producing superbly crafted objects for a few privileged lords and landowners. Apprentices learn the tricks of the trade for many years before developing the skills they need to make a livelihood on their own. The quality of workmanship is superb and each object is a highly customized, handcrafted masterpiece ­ no two are ever alike. But because so few objects can be handmade to this level of perfection, the cost restricts ownership to only a few.

This system remained essentially unchanged until the 17th century, when England's economy was based on cottage industries. Workers would purchase raw goods from a marketplace and individually produce finished goods at their homes, or cottages. The Cottage industries still suffered from low worker productivity and a snail's pace of production. Goods were still high in price and exclusive only to the wealthiest people.

Ultimately, the need for efficiency and speed forced this system to change. In 1733, the demand for cotton cloth soared, but limited production capacity threatened England's economic might. The flying shuttle, invented by a weaver named John Kay, cut weaving time in half. Though not readily accepted at first (many inventions of his time were deliberately destroyed by the people who thought they were going to be displaced by machines), new inventions such as the spinning jenny and the spinning mule, the water-powered frame, the power loom, and the cotton gin all improved the manufacture of cotton goods by speeding up the process. The factory system of mass production made formerly expensive, handmade items (such as shoes and gloves) more affordable to lowe-class and less wealthy people ­ many of whom now earned their wages working in other factories. The quality and productivity of worker's lives improved dramatically.

Interchangeable Parts and the Assembly Line
In 1907, Henry Ford announced that the Ford Motor Company's goal was to create "a motor car for the great multitude." Before Ford's time, automobiles were expensive for the same reason that Lord Dordogne's wedding gift was ­ the car was custom-made by craftspeople skilled in producing just one part while other highly trained people assembled the parts by hand into the finished product. Ford looked for more efficient ways to produce his car in order to lower its price and make it more available to the masses.

His first principle was interchangeable parts. Ford required that each individual car part be made the same way every time. Now any valve would be able to fit any engine, any steering column would fit any chassis. Eventually, by producing a common specification for each part, other companies could develop parts that would instantly work between and within the assemblies of others. Ford improved the machinery and tools used to make these interchangeable parts so a lower skilled worker could operate them, replacing the skilled craftsperson that formerly painstakingly made each part by hand, providing jobs to legions ­ and soon unions ­ of workers.

Another Ford idea came together in Michigan in 1913 when the first moving assembly line for large-scale manufacturing was introduced. With this assembly process ­ inspired by grain-mill conveyor belts and the meat-packing houses of Chicago that Ford had visited ­ work moved to the workers rather than the other way around. Ford trained each worker to manually perform just one of the 84 steps required to assemble the Model T and arranged the work so that as one task was finished, another began with minimum delay.

These innovations allowed Ford to produce cars at a much higher rate than his competitors ­ which led to lower prices and higher profits through increased capacity. For the first time, complex products such as automobiles could be produced with good quality, in a much shorter amount of time and for a reasonable price. The only major drawback to Ford's process was his "one-size-fits-all" approach. Ford joked that his customers could have their Model T in "any color they wanted, as long as it was black." Amazingly, the Model T ­ first built in 1908 ­ kept the original design until the last one ­ the 15 millionth - rolled off the assembly line nearly 20 years later.

Dynamic Assembly Brings 'Just-in-Time' Mass-Customized Production
It wasn't until the latter part of the 20th century that mass customization ­ factory-built yet custom-made products ­ was eventually achieved. It started with a new invention Ford could never have imagined; computer-controlled robots operating on dynamic assembly lines. These robots could be programmed to perform highly repetitive or dangerous tasks with much greater precision than the humans they replaced, and to do so around the clock.

Technology companies like Cisco, Dell, and Apple now build many of their most profitable products only when a customer requests it. Such requests are received electronically ­ invariably at the firm's Internet e-commerce site ­ culminating in an electronic command that is dispatched to a dynamic assembly production line to build each unique product. "Turns" of inventory are now measured in minutes instead of weeks or months, and thousands of variations of high-quality products are the norm. Dynamic assembly of products on demand promises to eliminate overproduction of products that sit unsold on shelves, further reducing the cost of goods. So today's modern factories are a natural evolution of Ford's factories where "just-in-time" production techniques and fully automated factory floors dynamically assemble individual component parts into a multitude of customized products on demand.

An important byproduct of this dynamic assembly approach has recently emerged; complex products such as aircrafts are now designed and tested as digital prototypes entirely in the memory circuits of a computer. Each part ­ from the fuselage to the bolts that hold down each seat ­ is precisely designed, modified, tested and redesigned to near-perfection before any actual building occurs. The optimized digital information is finally transmitted to computerized plants for manufacturing and dynamic assembly ­ where the finished product exists for the first time in the physical world. Today's products are designed and built using software tools that act much like spreadsheets in the financial world. "What-if" analysis, thermal and stress analysis, and cost/weight/safety tradeoffs are all done automatically by propagating even minor changes across thousands of related drawings. "Boutique" production runs are now possible in less than 24 hours.

But What Does All This Have to Do With Software?
Software has been silently following the same progression as the world of physical products. Like the artisans of old, software craftsmen would labor for years ­ sometimes in teams, sometimes alone ­ to create a single piece of software, their own work of art. Craftsmen like Dan Bricklin (VisiCalc), or Steve Jobs (MacOS) or Mark Andreeson (Mosaic/Netscape) would spend significant portions of their lives creating and then improving upon their software masterpieces. The search for the "killer app" became the preoccupation of users and investors. Software prowess was measured in millions of lines of code and the number years between major releases. An industry sprung up overnight around the need to produce better hand tools for developers ­ or to fill the holes in other software products. New companies emerged to produce compilers, spell checkers, debuggers, virus protection, and security products to name a few. This cottage industry still exists today.

Yet for more than 20 years we've quietly been in the second phase of the software industrial revolution where object-oriented programming and interchangeable parts ­ objects, "beans", components, and libraries with standardized interfaces ­ have become the norm. In many ways this trend is analogous to the concept of inventories of "off-the-shelf" interchangeable parts that people could use to build higher order assemblies. Like the experts focused on just one step in the assembly of the Model T, teams of programmers divide the application development workload, focus on their own competencies and manually assemble a variety of products from a known set of parts. But like those same factories of old, the existence of a huge inventory of parts doesn't solve the problem of building customized versions of each application for the needs of an ever more disparate and demanding audience. The lack of an automated, mass-production process has made economy of scale for Web applications an elusive goal.

Today's Challenges
So today's developers are swamped with demands to build a new type of application ­ using yesterday's tools. They need to build multiple, complex Web applications that provide access to business processes and data contained in various backend systems. In many cases developers have already deployed dozens of Web applications and now find they are unable to customize new application variations for various audiences and channels. Keeping up with the rampant demands of continuous change in business requirements is a constant battle. Suddenly, the need for efficiency and speed is forcing traditional approaches to Web application development, deployment and maintenance to change.

The Future of Web Services: Dynamic Assembly
So the stage is set for software's third phase, the dynamic assembly of complex applications on demand. The recent emergence of Web services ­ Web-accessible software applications with standardized or self-describing interfaces ­ brings new possibilities and new complexities to assembling Web applications from an ever-increasing supply of disparate component parts.

The need to assemble pieces of an application at runtime isn't new; the notion of breaking applications into pieces that get loaded together at runtime is common. WebLogic Application Server performs such tasks on EJBs and loosely coupled Web services in its runtime execution environment. But what is now required is an ability to make choices during the assembly process about which components should be used and how to customize the application's structure, functionality, behavior, and content out of such components. How can this feat be accomplished?

Developers have used many combinations of explicit instructions coupled with rules engines to build the appropriate assembly instructions for different use cases. But because developers have to manually construct all the variations of the rules, templates, and other pieces that drive the dynamic assembly process, they quickly get bogged down in maintaining a morass of hard-wired connections, if-then-else statements, and other explicit code-glue to make it all work.

The introduction of programmable robots on the modern assembly line enabled mass customization and incredibly short design-to-manufacture cycle times. Software factories, which mirror their physical world counterparts, have recently emerged that capture the application design process into discrete steps and use software robots to dynamically assemble components and Web services into highly customized composite Web applications on demand.

Bowstreet's aptly named Business Web Factory is a Web services development and assembly platform that automates the creation and maintenance of complex Web applications. Patent-pending robotic software agents called Builders enable programming tasks to be captured in software (for instance, generating the code that calls a SOAP/WSDL Web service). At runtime, specialized instructions invoke a sequence of Builders, who perform their various construction tasks during the "regeneration" of the Web application. Since that of the execution of the Builders during regeneration can be varied (much like computerized robots on a modern factory floor) through parametric inputs contained in profiles, this approach automatically generates a wide range of different Web application variations.

Business Web Factory routinely generates hundreds or even thousands of application variations, resulting in significant timesaving over successive development projects, and allows business users or the end users themselves to "order" new applications to be built on demand simply by creating new profiles. Like a modern factory, this approach reduces the need to maintain legions of separate applications (products sitting unsold on shelves), builds a call to a Web service (parts) on demand, and eliminates repetitive or boring programming tasks (robots) that take up too much development time and defocus the programmer from more important, revenue-producing development tasks.

This approach also inherits some of the same important benefits we saw in the CAD/CAM systems that drive today's modern factories. A single change can ripple instantly through tens, hundreds, or even thousands of applications ­ and because there aren't separate applications to constantly change, update and fix, the time and cost of application maintenance is greatly reduced. The developer is thus relieved of tedious, repetitive work associated with maintaining separate code bases while eliminating expensive regression testing. And for the first time, business people can control applications and respond to changes quickly and independently, allowing them to concentrate on creating competitive advantage for their firms, rather than helping IT manage a sprawling Web application franchise. "Boutique" production runs of highly customized and/or specialized Web applications can be done instantly.

So from the exquisite, expensive, one-of-a-kind products laboriously produced by an artisan to today's highly customized, high-quality products dynamically assembled at low cost and high speed (Figure 1), the software industry has closely followed the trend of the Industrial Revolution. We fully expect that history will record that the solution to the enormous cost and complexity of application development, deployment, and maintenance will be seen as having started with the dynamic assembly of Web services.