The Cure for Silver Bullet Syndrome

I had an invigorating conversation with a co-worker the other day. We were discussing the next generation of technologies for application component integration: Web Services and SOAA. As we got into the discussion, he made a statement to the effect that tools will help flatten out any difficult learning curve in implementing the architecture. This seems to be a fundamental flaw in thought at businesses worldwide, and thus is a worthy topic for consideration here.

What is it about software "tools" that allure? In my experience as a software engineer I have a great number of "tools" that I have had exposure to to "help" in the development process. Unfortunately, precious few actually were helpful, and more than a couple were complete disasters.

There was one tool, called Xenos d2e which was supposed to help our company mine print data so we could reformat the document with a new "look and feel" that conformed to the client's changing taste. This tool worked well enough -- given the page were static. Add too much dynamic formatting and the tool would loose track of its page markers. Of course that could be worked around by scripting code on the back end -- in REXX. Well, nobody knew REXX, so those tasked to work with this product had to ramp up quickly on the language. Then when the project transitioned to a new team the new team had to learn REXX to maintain the software. The tool was unhelpful.

Contrast that experience with a tool written by a colleague which did almost the same thing, though it was written without a back-end programming language. The tool did exactly what we needed -- to extract data from the input print stream based on defined markers. The difference? Even though the tool written by my colleague was less capable, it was designed to do exactly what was needed, took less overall time to learn and had fewer maintenance issues. This tool was helpful, and cost less on learning, maintenance and license fees.

In my current position, I deal with a certain source control package which has many integrated features and lots of bells and whistles. Unfortunately it cannot easily define multiple lines of development against a common code base, and therefore we have much difficulty in parallel development for small enhancements based on current production while doing major feature rewrites for upcoming releases. This extra management costs quite a bit of time and therefore is not efficient. And yet we gladly paid $20,000 for the privilege of being less productive than paying $0 and using Subversion, but lacking an integrated bug tracking function that we rarely make use of anyway.

Kathy Sierra over at Creating Passionate Users foresees an even more sinister issue. By hiding the implementation details, Kathy argues that we are training ourselves to not understand how technology works. One initial reaction might be that since the tool helps you get the job you want done, you don't need to know the underlying technology. This might be true for many end users, but it's extremely important for technology service providers to understand what's going on underneath the surface.

What were to happen if you suddenly had a client that was using a slightly different version of the technology? For instance, if your tool handled web services abstraction reasonably well, but your next 800 lb gorilla client used a web services handler from IBM that deviated from the standard in subtle but frustrating ways and your tool would be worthless.* If you never bothered to learn how to implement a web service without the tool, you would be stuck. Likewise if the spec was revised by the standards body and your tool's author had let the tool become obsolete. It's quite likely that your tool would limit your effectiveness in dealing with clients that had recently purchased their technology stack.

Don't let it happen to you! Use tools, sure. But be sure you know what's beneath the surface.

* I'm not saying that IBM has a deviant web services stack, but they're an easy target as their J2EE stack seems to deviate from the spec.