identify good engineer in this AI Fog !!!

Previously people were hired particularly in tech companies based on their muscle memory to remember DSA algorithms, and CS basics concepts, though these are good enough to distinguish genuine allies from opportunistic ones. But now the ceiling is high and we have to redesign the style of interview so that it becomes the funnel to get good software engineers for our companies or organization, that dont just boast on their CS skills and leetcode numbers, but the ones that actually help to generate more ROI for your company and handle the complete SDLC by itself, cutting the cost also, for this i was thinking the best funnel to design, then i came across one of the interviews of Andrej Karpathy, whom I would describe as an engineer who helped shape modern AI at OpenAI and Tesla, he suggested a very beneficial sort of criteria to hire the best engineer.
According to him, to any new engineer in the interview process, we should give him complete access to agentic tools with modern AI, and instead of asking him to write an algorithm of lets say 100 lines, we tell him to build some big system let say X (former Twitter), not full fledge, but at least all the basic features, and let him use all the knowledge resources to design that app, after he completed the application then the interviewer use many agents parallely to test this application , across stress tolerance, availability, security vulnerabilities, and scalability, and try to break his application.

Only if the application survive or atleast don't break completely (obviously there will be some room for improvement), then the engineer will be worth considering
And for me, this is the best way to sort out the good engineers from delusional engineers, because

You can outsource your thinking but you can't outsource your understanding