As someone who scoffs at "AI" and a professional EE who detests autorouters (they take more time to set up than it takes to do the job), I honestly think this is the future. They're a good match for each other, since routing has clear success and failure conditions, a lot of leeway to achieve them (actually too much; this is what usually sinks autorouters, with them painting themselves into dumb corners), no expectation of or need for perfection, and the default presence of a human in the loop (to add the logos if nothing else).
But the real payoff will be in automatic component placement, since that's 85% of the work anyway. And I think the same trends apply there too, it will just take longer.
> But the real payoff will be in automatic component placement, since that's 85% of the work anyway.
I wonder how much separable the problems of placement and routing are. That said, I'd love to see a function in KiCad that does auto-placement (more than I want auto-routing).
In terms of fully optimising the final result, they aren't separable. If you want routing to be likely to succeed, you need to give extra space between components, but this will waste some of it. Getting a good result means the placer needs to have at least some knowledge of the routing. This tradeoff also exists with manual placement and routing to some extent, especially since the tooling it still not very good at push-and-shove for a whole component, especially one with many pins. Also in practice in the design process there's quite often no requirement for an optimal result in terms of space used: before the PCB is designed there's a negotiation with the mechanical design as to how much space is allocated and so long as you're under that you're OK, and there's little bonus points for coming under (since it would require reworking the mechanical design to take advantage of it).
But the real payoff will be in automatic component placement, since that's 85% of the work anyway. And I think the same trends apply there too, it will just take longer.