I worry about these remediation measures. The problem described is that the increased reliance on AI has increased cognitive debt, and the solution is to write tests that better capture intent and to update design documents continuously. But the people with cognitive debt problems that got there by using AI are also going to be using AI to write tests that "capture intent better" and using AI to update design documents. And while, sure, those are good things for your agents to be doing just for the sake of the effectiveness of those same agents, it's also not going to help you get out of the cognitive debt spiral.