I did some more studying on boids, and I've had some revisiting to Unity3D. Especially now that they're about to port the unity editor for linux.
Boids have been themselves explained very well elsewhere. The meat is in combining several simple behaviors to produce emergent complex behavior such as moving in an unstructured flock.
In-genera, there are some pitfalls to avoid at implementing these things for constantly moving agents. It is easy to implement individual behaviors so that they brake the unit. Care must be taken to ensure that each steering behavior does exactly what it's been programmed to do. Wrong implementation results in only slightly wrong behavior so it's hard to detect.
Although the boids algorithm forms convincing flocks, individual units in the flock may do nonsensible decisions. Therefore it's not as useful in its usual form as I've thought. But it appears as a good stepping stone for constructing convincing computer game AI. For example, you can find this article about context behaviors that seem to improve on boids algorithm to make it more fitting for simulating behavior in small groups by plotting the behaviors into danger and interest maps to form the decision instead of summing them all together.
It can be I'll continue with some more game related topics in the blog. I don't feel like I'd be done yet here. The elm-bonsai and pyllisp got to wait a bit.