Super hard problems are often solved by making strange weird connections derived from deep experience plus luck. Like finding the one right key in a pile of keys. The intuition you used to solve your problem IS probably beyond current agents. But, that too will change perhaps by harnessing the penchant of these systems to “hallucinate”? Or, some method or separate algorithm for dealing with super hard problems creatively and systematically. Recently, I was working on a hard imaging problem (for me) and remembered a bug I had inadvertently introduced and fixed a few days earlier. I was like wait a minute - because in that random bug I saw opportunity and was able to actually use the bug to solve my problem. I went back to my agent and it agreed that there was virtually no way it could have ever seen and solved the problem in that way. But that too will come. Rest assured.
PathGetter really helps me a lot. Drag n droplet that puts any items path in memory. That’s just one example. Essentially I have several “personal” programming projects going at all times. I’m working on them all the time. Most progress is made away from the keyboard while walking or even sleeping.