Can Agile work for a team of experts?
brendel.com
brendel.com
A project manager has many roles, but one of the most important is to figure out which bits of the chosen methodology do apply, and which don't. There is no such thing as a "best practice" - only a "good practice within a certain context". The PM's job is to figure out which practices match the current context, and monitor them as the context changes and certain practices drop out (and others drop in).
What you end up with, if you have a good PM, is a unique methodology tailored to your specific project, which is the only sensible thing, since all combinations of project and project team are, in fact, pretty much unique.
It's worth adding that the PM doesn't need to be full-time - but I'd say that even on a 2-3 people start-up dev team you need at least one person who's experienced enough to take on that role.
Each user feature probably requires spiking through the expertise of all 4 devs. If people don't work together, APIs and workflows can diverge and then integration becomes horrible. So everyone (+client) has to get together to build acceptance tests and integration tests between different layers of the framework.
And I think the article is wrong - pair programming does work here, but not on domain-specific issues. The most useful way is to pair program the interfaces between layers/silos.
The database guy and the kernel guy should pair program the kernel module that spits data out, and the DB module that loads its up. This gives each person a feel for the other's assumptions, and lets them lay out a smarter interface that takes advantage of both devs' silo'd domain knowledge.
Also, even if you don't do cross-silo pair programming, code reviews are still useful. This can be totally informal, but saying "Hey can I walk you through my code?" discovers a lot of bugs, stupid misunderstandings, etc. Doing it frequently (daily) is useful because it makes you code with the understanding that you'll have to show it off later to someone who won't get it unless it's really clear - so the code stays clean, thus (hopefully) better. Once again, the bonus is that a common metaphor emerges even among more disparate members of the team.
The point of Agile is that it's agile. As in, nimble. Flexible. Suited to your requirements. Subject to your needs and intelligent application.
I'm curious as to how other people deal with this.
A good methodology is not like SQL programming. The system doesn't explode if you don't follow every rule to the letter. In the immortal words of R. A. Heinlein, a methodology is an art that you pursue rather than a goal that you expect to achieve.
A methodology is like playing Dungeons and Dragons. There are rules. You gather together, you play the game, and you generally follow the rules. But the DM is empowered to bend or break any specific rule if it makes the story better or brings the game closer to the goal.
Many RPG designers have suggested that you don't even need the rules. You could just gather everyone in a room and say "okay, let's all pretend that we are elves and dwarves". And that can work. The very best storytellers can get away with this. But, in general, people prefer games with rules, even if they're very sparse and loose rules. They provide important structure, and they give you an excuse for saying or doing things that other people might not like.