Today we had a discussion about which agile practices work, and why some tasks just seem to take longer than necessary. One of the reasons seems to be that we don’t really reuse concepts and components very much. Every scrum team has responsibility for meeting their goals and satisfying the customers, but nobody is focused on maintaining a somewhat common architecture, and practices, both good and bad, are not communicated reliably to other teams.
This is not necessarily bad. Most of the developers are pretty bright people, who actually work at companies like Trifork because that way they won’t get some Enterprise Architecture(tm) pushed down their throat. Instead, there is an expectation that everybody will strive to do their best for the project.
The whole agile culture empowers the developers – this is one of the core ideas. The developers get to speak directly with the customer, design solutions, implement them, test them, document them – generally the developers are responsible for as much of the development as possible. A good thing, but maybe developers sometimes have a little too much to say. I don’t particularly like the term architect (mostly because “whiteboard” should have been prepended), but having Scrum teams without a proper architect doesn’t really seem optimal in the long run. By proper architect I mean one who has the capacity to create consistency across projects, and take decisions about key components, languages, and so on. The architect shouldn’t necessarily make all the decisions, but some things are just not worth discussing again and again. Examples of this could be whether to use a coding standard or not, or whether to include a completely new and untested technology. Another classic seems to be that decisions are never final. This means that somebody might spend time in integrating a new tool or component, even though an earlier decision was made against it.
Granted, some of the decisions made on a project are just plain wrong, at least in hindsight, but it just seems that experimenting with new technologies on a project which is due to deliver a working product isn’t the best idea.
As usual, the problem comes down to communication. Scrum and XP handles team communication pretty nicely. We have daily scrums, iteration meetings, retrospectives, and so on, but there are no practices to establish inter-team-communication. You could, of course, go for something like CMMI, which has almost complete focus on organization-wide communication, but that just doesn’t sound very good (theoretically it should be possible, though. My master’s project was about exactly this). In other words, we need some solid agile practices for enabling communication across the organization.
I wish I had some good ideas as to how this should be done. Ask Jeff Sutherland, and he’ll probably begin talking about scrum of scrums, basically arranging teams hierarchically, and letting the scrum masters communicate between the teams. On the face of it, this looks like a good solution, but somehow I don’t quite like it. In really large organizations, it’s probably the only way, but is it appropriate when you have 50-100 developers?
My best proposal is to separate the day-to-day development from the talking and the experimenting. Development happens in scrum teams, and the teams live their own lives, but the design decisions are coordinated with what we could call architects. Maybe librarians would be a better title, because their main purpose is to know what others are doing and have done, what has worked, and what has not worked. The talking and experimenting would take place at other times. This could be the 20% time which Google employees have to spend on something else, it could be something like the JAOO meetups at Trifork, it could be video blogs. Basically anything, the important thing to somehow promote a culture where both learning and teaching is essential. Of course, this requires some ongoing investment from the organization, like the 20% time.
If this separation exists, we as developers might be able to focus better on creating actual value, and we might also become better at learning from each other. It’s important to notice that the separation is there to avoid too much distraction when developing. However, everything good and bad that comes out of the talking and experimenting should be taken into consideration when making important architectural decisions – it’s just a question of where the experimenting takes place, and when the results are brought back into a project. Most of the time, our projects are relatively short, so after committing to some architecture, it doesn’t really pay off to change it.
I expect some of this will be addressed, for example at future JAOO and QCon conferences. In the meantime, it would be nice to hear about inter-team communication patterns – how to ensure common knowledge, and how new ideas, technologies, patterns, principles, and so on can be distributed in an organization full of independent teams.