This is the part that all the autorouters seem to miss. A board design may be represented in the computer by a schematic, but in reality it's a schematic + a huge amount of engineer knowledge. The person laying out the PCB needs to be able to look at the schematic and know that the switching regulator has very precise layout requirements, but that status LED can be at the end of a long trace.
That information simply isn't captured in current schematic software. Until it is I can't see autorouters being effective.
I can see a world where every schematic includes simulation models, and the autorouter uses simulation data to know exactly what frequencies are moving down each net and in each location. That requires detailed spice models of every component on your circuit though - so it probably wouldn't save the designer any time anyway as they're just doing different work. I'm not even sure how you'd simulate the signals coming out of a microcontroller - how does the autorouter know that one PWM IO is producing a 500kHz clock into a high current switch and the other PWM IO is producing a fixed 3.3V?
So then maybe you need to incorporate not only simulation models, but your actual CPU code. Which then means you need high quality microcontroller and FPGA emulators. There's a new problem!
There's probably a middle ground - a designer could annotate each net with a waveform which would be fed into the spice simulation - but even then we're talking significantly more work than just laying it out yourself.