Firstly, I assume you know the purpose of route aggregation (if you need more clarification on that, feel free to post another question, as it is another topic by itself).
Secondly, in this context, what Tanenbaum is saying is that aggregation can be done manually, but it is done normally done automatically by the routing processes. Manual aggregation is when a human looks at the various network destinations and decides which ones can be aggregated, and then enters the aggregated network destination info into the router configuration (during the process where we have, as Tanenbaum writes it, "an administrator assigning addresses to networks"). The alternative is, for the routing processes to handle it automatically.
Yes, it can be turned on or off through the router configuration. In fact, it can be turned on or off separately for each routing protocol being run on the router (if more than one routing protocol is operating). It is perfectly fine to turn aggregation off for edge routers that are not connected to many networks. Then, their routing table will have individual entries for more networks (not aggregated), but would still not be too big. However, for routers in the Internet core, normally aggregation must be turned on so the routing tables don't become too big (200,000 prefixes is the size Tanenbaum mentions; without aggregation, they could be much bigger with many more prefixes).
It is an "automatic process" in the sense that the routers will exchange information as part of the routing protocol(s), and they can figure out the aggregation automatically, as part of their logic. "I mean isn't process especially on network is always automatic?" -> some things can also be configured manually. Hence, Tanenbaum tries to make the point that most aggregation is handled through the processes automatically.