I've been following the debate in the blogoshere that was ignited by the less-is-more movement spearheaded by 37signals for some time trying to think of how to write the first of many blog posts about it. On one side are all the folks who (like me) felt that a lot of heavy-weight development techniques were being applied to web development (functional specs, loads of QA, explicit design vs. development phases, a bunch of efforts to reinvent most of the LAMP infrastructure with every app, to name a few) and that despite being "professional" these things were actually hurting rather helping the cause (building useful and popular webapps). On the other side are folks who claim that software development is software development and "real" products will always require these "formal" support structures.
Today I had a work-related email conversation which made me think of this again, and then tonight while catching up on my feeds I came across a presentation which helped me get over a mental hiccup brought on by the email.
Let me start at the beginning though. This morning while writing on our corporate wiki about our current project status I wrote an offhand comment:
In web speak, with a small and agile team as talented as this one, 12 weeks ought to be enough time to build almost any product.
This landed an email from a co-worker who basically called me out for making such a crazy generalization. At first, I felt like maybe I had drunk a little too much of the 37signals kool-aid with respect to speed of development (It should be noted that while I like what they have to say in principle, in practice I think it should be taken with a grain of salt given that most of what they put out there are very thin web frontends to database-driven workflow apps). Of course, I wrote back right away with the reasons why I felt that a generalization like this one can hold, namely that in the webapp-as-service world:
- You don't ship the bits to the customer which means you can piece it together in the back end and rely on real-time patching to fix things.
- Browser apps are better for doing incrementally interface-wise so you can draw the big blocks up there and then refine in a way that you just can't when you are running a fat client app.
- The general shape of web apps (stateless protocol that hits and app server which uses a DB and a filesystem) is relatively straight forward (hence low overall complexity). The real challenge comes (in my opinion) in User Experience and scalability.
I feel pretty good about why these three things cut down on the time to build web applications but still felt a little weird about making a generalization in the first place. But then I heard this Salesforce.com conference by Adam Bosworth (see sidebar) about the development model at Google and something clicked for me. It's not that every web app can necessarily be built in three months (though I still hold that most can for the reasons listed above), it's that all of them should be. Because unlike the era of productivity applications when once the bits shipped the app was complete until the next release (except for bugfixes), most interesting web applications are community applications which necessarily means that when they "ship" they are only half done at best. The other half comes from looking at the logs, reading the customer email, finding what works and what doesn't and managing this iteration cycle to be as tight as is humanly possible (I can't remember when I heard that at Flickr this could be as short as 1 hour.
So, go listen to Bosworth (it's short). We're in a new world and I for one am thankful we have the right tools to meet the ever tightening development schedules.