by Leon Rosenshein

How Buildings Learn

Now that I’ve written about Seeing Like A State, I want to talk about How Buildings Learn. How Buildings Learn is, in many ways, a counterpoint to Seeing Like a State. It also has a lot of relevance to software design.

How Buildings Learn starts from the premise that over-architecting is bad. That the best way ensure longevity is to architect not only for the now, but also for the future. And then, when you get feedback, listen to it and adapt. It’s a very agile way of approaching building architecture.

Brand goes into some detail about how designing for specific constraints is limiting. Of course, that makes sense. Every time you optimize of one thing, you’re not optimizing for something else. Instead, what Brand recommends is simple designs that you can adjust as you learn the true usage patterns.

There are multiple examples where premature optimization of architecture has caused problems. Consider the Fuller Dome. If all you’re trying to do is minimize resource usage it’s a great idea. Or if you’re building in a zero gravity environment. If you’re not, then you end up with a lot of wasted space in the top/center of the dome. Other examples are Falling Water and Villa Savoye. Both are examples of form trumping function and causing problems later on.

Both construction and software use the term architecture, but does Brand’s approach really apply? After all, the Unix Way is all about being specific. Doing one thing and doing it well. Which is the opposite of what Brand proposes. Or is it? The Unix way is not just about doing one thing. It’s also about composability. Which is really what Brand is getting at. Build something that is easy to subdivide into parts that are composable. Build the parts that meet your current need. When things change or you learn you need more, adjust the parts to match the new understanding. That’s evolutionary architecture.

And that’s in direct contrast to pre-defined legibility. A state (or organization) is often looking for control and predictability. So instead of building something that could work and them adjusting it to fit the exact needs, it asks for detailed, involved, plans. And then it sticks with them, even if the reality on the ground shows problems (see Villa Savoye above).

Another, more software based approach, but still with an architectural basis, is The Cathedral and the Bazzar. In it, Raymond describes the differences between working from a centrally defined/controlled plan and working from a set of common goals. According to Raymond, the Bazaar will get you a superior result. He’s got more than a little evidence to prove it.

However, the model of starting with something adaptable and a set of common goals and then building the perfect building (or piece of software or really any other shared resource) comes with its own problems. Not the least of which is diffusion of responsibility. How you handle that issue is critical to having a good outcome when buildings (or code) learn. Anarchy is not the way to reach a solution that optimizes for what everyone is trying to get done.

Which leads right to Ostrom’s Governing the Commons. But that’s a topic for another day.