Erm, that's my opposite opinion, I guess. My (attempted) argument is that one should not write their own framework unless doing so as an exercise in learning. In any other case, using existing frameworks or using the "roll your own library mix tape" strategy is far more efficient.
Basically, I'm saying if you're using a language which has mature frameworks and/or mature libraries which could be forged into an ad-hoc framework (templating, serving, ORM, etc.) it is terrible advise to tell people "just write your own".
Only in the case that you've considered all existing (relevant) options and determined none of them meet your use case (and none can be minimally altered to do so) should you consider writing your own. Anything else is a waste of time and effort.
Right, your new rule is, write you own framework/library if necessary, otherwise use what's there. That's much better than your original rule which ruled out any use of frameworks as no frameworks could ever have been created.
But Im' still not happy :) I'm sure a lot of popular frameworks were created without any necessity, simply because they were fun to make. Some turned out to have qualities that nobody (including its creator) would have been able to spell out beforehand.
Basically, I'm saying if you're using a language which has mature frameworks and/or mature libraries which could be forged into an ad-hoc framework (templating, serving, ORM, etc.) it is terrible advise to tell people "just write your own".
Only in the case that you've considered all existing (relevant) options and determined none of them meet your use case (and none can be minimally altered to do so) should you consider writing your own. Anything else is a waste of time and effort.