When we first introduced Papert's principle — that is, the idea of growing by inserting new levels of management into old agencies — we did not ask when new layers should be built. If managers are inserted too soon, when their workers are still immature, little will get done. And if those managers come too late, that, too, would delay the mental growth. What could ensure that managers are not engaged too late or too soon? We all know children who seem to have matured too quickly or too slowly, in various respects, to match other areas of their growth. In an ideal system, each developing agency would be controlled by another agency equipped to introduce new agents just when they're needed — that is, when enough has been learned to justify the start of another stage. In any case, it would surely be disastrous if all our potential capacity to learn became available too soon. If every agent could learn from birth, they'd all be overwhelmed by infantile ideas.
One way to regulate such things would be to actuate new agencies at genetically predetermined times. At various stages of biological maturity, certain classes of agents would be enabled to establish new connections, while others would be forced to slow their growth by making permanent connections that, till then, had been reversible. Could any clockwork scheme like this be guaranteed to work? Consider the fact that most of our children acquire agents like Reversible and Confined before they are five years old. For those children, at least, it would suffice to activate new intermediate-level agents at that age, so those children could proceed to build agents like Appearance and History. However, children who weren't ready yet would then be slightly handicapped by being forced to build some less-than-usually effective Societies-of-More. Nor would that rigid maturation- schedule serve well those children who had already moved ahead of schedule. It would be better to have systems in which the timing of each stage depends on what has actually happened earlier.
One way a stagelike episode might start could stem from what we called the investment principle: once a certain skill surpasses all its close competitors, it becomes increasingly likely to be employed — and thereby increases its opportunities to develop even further. This self-enhancing effect can cause a spurt of rapid progress in which a particular skill quickly comes to dominate the scene. One way a stagelike episode might end could stem from what we called the exception principle. To see how this could happen, suppose that a certain agency develops so useful a way to do some job that many other agencies soon learn to exploit that capability. The more those other agencies become dependent on that skill, the more disruption will result from every further improvement in it — since it now has more customers to please! Even increasing the speed of one process could damage other agencies that depend upon how long it takes to work. Thus, once a scheme persists for long enough, it gets to be extremely hard to change — not because of limitations inherent in itself or in the agency that developed it, but because of how the rest of the society depends upon its present form.
Once it becomes too hard to change an old agency, it is time to build another one; further progress may require revolution rather than evolution. This is another reason why a complex system must be grown in a sequence of separate steps.