Actually, he did.   But just like us, Satoshi didn't have any crystal ball, and thus no way to objectively determine what the block limit should actually be.  We can take an average of the transactions over the past four years, but that would not account for either protocol enforcible contracts nor transaction scripting that is likely to eventually increase the average size of transactions.  Nor would it account for many-in-many-out transactions that would only become common in the context of some large companies' weekly payroll.  The problem is that we don't know what the future holds, and can only make educated guesses as to how bitcoin will be used in the future.  There is no precedent for any of this guys, try to remember that we are all pioneers in this field and no one really has any applicable experience to guide us.

Satoshi chose a hard limit of 1 meg as an arbitrary design decision, effectively putting off the issue till the future, wherein hopefully there would be more useful data to make a more informed decision.  Unfortunately, we really don't have better data; at least not data that we know how to use for this purpose.  And since Satoshi left, we really don't have a tiebreaker to make a final decision when we can't come to a consensus.  Thus, what we are left with is some measure of dissent no matter what we choose to do.  The soft limit really was a short term convention.  Can we come to a decision regarding that first?  That is actually more pressing.  Should we permit the miners to double that soft limit, comment out the soft limit altogether, or include passive/aggressive enforcement rules to the vanilla clients in order to punish violators, or nothing at all.  Bear in mind, nothing is the default condition and that is what we will get with gridlock.