Years ago the long-gone and unlamented OrCAD Layout had a spreadsheet view of your nets that was a slightly better than good-enough interface for setting constraints for autorouting. Once you got your footprints, placement, constraints, and hand-routed nets locked down, you could iterate very quickly.
It's nice to see anyone at all working on PCB autorouters, which have been pretty stagnant since Cadence bought SPECCTRA in the nineties. The guys who wrote SPECCTRA went to the VLSI world iirc and never came back - I guess that's where all the fame and fortune is. Probably was a patent minefield for a while there too; might still be. Autoplacement was an utterly intractable problem back then, still seems to be, and it's probably ripe for a generative AI approach - a good generative AI first pass at parts placement could be a net time saver. The biggest problem would be convincing the cranks that you can be good even if you can't be perfect.
I'm sort of puzzled by the kids out there trying to do schematics-as-code. On the one hand, I would love to see that work as a backend format - especially where folks like the jitx guys seem to be making a lot of progress in encoding appnote and datasheet-level design rules into parts models. Reading every datasheet in the detail needed for commercial designs is more work than people expect, and so is getting junior engineers on board with the process. So any automation there is beneficial. The problem is that the approaches all seem rooted in this idea of schematics as data entry for layout, a kind of source code, rather than as design documentation with a carefully evolved visual language that needs to be accessible to people who won't have an EDA suite installed on their computer. I guess people who learned from puzzling out schematics in the Adafruit/Sparkfun/Shenzhen style with explicit wiring minimized might not understand the value of a good schematic.
The other thing is leaning too far into thinking-by-analogy and trying to make PCB-level design like VLSI design. I don't think it's entirely impossible - if we had better tools for DRC and validation, component-level design would look more like VLSI design. But the design space is so different, with such loose coupling between design, EDA/CAM/simulation, validation, fabricators, assemblers, component vendors, regulators/certifiers, and so on that just getting one corner of it really right would be a major accomplishment.
It's nice to see anyone at all working on PCB autorouters, which have been pretty stagnant since Cadence bought SPECCTRA in the nineties. The guys who wrote SPECCTRA went to the VLSI world iirc and never came back - I guess that's where all the fame and fortune is. Probably was a patent minefield for a while there too; might still be. Autoplacement was an utterly intractable problem back then, still seems to be, and it's probably ripe for a generative AI approach - a good generative AI first pass at parts placement could be a net time saver. The biggest problem would be convincing the cranks that you can be good even if you can't be perfect.
I'm sort of puzzled by the kids out there trying to do schematics-as-code. On the one hand, I would love to see that work as a backend format - especially where folks like the jitx guys seem to be making a lot of progress in encoding appnote and datasheet-level design rules into parts models. Reading every datasheet in the detail needed for commercial designs is more work than people expect, and so is getting junior engineers on board with the process. So any automation there is beneficial. The problem is that the approaches all seem rooted in this idea of schematics as data entry for layout, a kind of source code, rather than as design documentation with a carefully evolved visual language that needs to be accessible to people who won't have an EDA suite installed on their computer. I guess people who learned from puzzling out schematics in the Adafruit/Sparkfun/Shenzhen style with explicit wiring minimized might not understand the value of a good schematic.
The other thing is leaning too far into thinking-by-analogy and trying to make PCB-level design like VLSI design. I don't think it's entirely impossible - if we had better tools for DRC and validation, component-level design would look more like VLSI design. But the design space is so different, with such loose coupling between design, EDA/CAM/simulation, validation, fabricators, assemblers, component vendors, regulators/certifiers, and so on that just getting one corner of it really right would be a major accomplishment.