I maintain that specialization is crucial for software development, and that overgeneralization can be just as harmful as overspecialization. The following figure will illustrate my point.
At the top of the figure is a team of three with each having a specialty skill. Each team member happily sits in their own comfort zone and there is very little understanding of adjacent specialities. Since finishing an item of work needs all three specializations there will be handoffs and theory of constraints applies. Even if the team is working on more than one item time is wasted on waiting for bottleneck resources and so on. The good news is that since there is a lot of expertise in the team the finished product will reflect that expertise (if it ever gets finished).
Now let's look at the other extreme at the bottom of the figure. A team with nothing but generalists will work most efficiently since everyone is able to do everyone else's job. Very little time is wasted since there are no handoffs and everyone is able to support each other. The caveat is that since everyone is a generalist the team lacks mastery in any field, and thefore can build mediocre products at best. Jack of all trades, master of none applies.
In summary pure specialists produce excellent products late and pure generalists sub-standard products on-time.
How about the middle road then, the "generalizing specialists"? The guys who are specialists in their field, but know enough of their surroundings to be able to work effectively with other specialists. Perhaps they have the social and teamworking skills to also pick up new skills and broaden their specialty. Such a team should be able to work efficiently and still utilize the expertise of individuals to build good products.
A mixture of specialization and generalization is needed, though I will grant that overspecialization may be a bigger problem in the industry.
By the way, how does generalization work in a truly cross-functional team that includes other disciplines besides software development? Is it feasible to train graphic designers software design, and vice-versa? What about marketing and sales? In the long run it probably is beneficial to send all developers on a marketing crash-course but in a project environment I wonder if it is feasible?
