Max block size could be retargeted periodically alongside difficulty adjustments using average block size and the frequency of full blocks in a period with the hard-coded value as a floor.

Imagining the necessity of 20MB blocks relies on the assumption that a massive increase in transaction volume develops, but what if it slows significantly?  Bloated block broadcast delay might temporarily even be useful as a competitive advantage.  When fees outweigh reward, doesn't the mining market encourage bloat?

Also, with a hard limit on size, won't the conversation on manual adjustment resurface indefinitely with increasing political difficulty?