As a separate note, but tied to Ward's article, it's been eye-opening to me to finally understand some of the motivations for existing configuration management (CM) processes.

One thing that's become clearer to me in recent years is that management is hung up on rigid CM processes because the cost of errors has been high. And further, the cost of experimentation in other engineering fields is high. For example, you can breadboard a circuit but that's a long ways from production. And the problem of getting from prototype to production is expensive for hardware, plus mistakes are even more expensive.

With software, there's a change. The cost of mistakes doesn't have to be high (unless we make it high). And the cost of experimentation is relatively low, compared to other engineering fields.

One thing I've been trying to sell to some management people I know is the idea that CM for software is fundamentally different than CM for hardware. They don't want to buy the idea. But more and more, I am becoming convinced that software is a different thing, and that ideas like collective ownership are going to slowly expose this radical difference.