I was waiting for someone to bring up the QM2k. The QM2k plus the CT digital scan amps basically equals what I was talking about- except in my world, you'd remove the analogue link between the (digital) scan amps and the (digital) QM2k, and feed the galvo feedback sensors right back into the QM2k. Then, you make the QM2k really really small (easy to do nowadays- the Pangolin hardware designs are pretty old by microcontroller standards) and build it into the scanner amp. Preferably, you'd also have closed-loop control of your laser drivers or modulators too. I don't really care whether it's analogue or digital at that point since their response is so much simpler than galvos.
At this point it makes sense to start describing things in terms of visual elements- and we are not talking about a complex description here, just saying circles, curves, lines and dots- rather than as line segments, because if the system knows which lines are supposed to approximate curves and which are supposed to approximate straight lines, it can make better optimizations.
I don't see what the big problem with doing this is. I remember there being similar arguments made against compact discs and even colour TV.