Everyone knows about tech debt. You cut corners on code quality to ship faster, and you pay for it later. We're definitely watching a new version of that emerge in real time, except instead of deferring manageable code, you're deferring actual understanding. And unlike tech debt, cognitive debt compounds invisibly. You don't get a failing test suite. You just get someone who can't debug their own project, can't evaluate whether the AI's suggestion is good, and can't extend what they've built without prompting their way through it again. What I keep thinking about is where this leads at scale. Right now it's mostly developers vibe-coding their way through projects they half-understand. But AI is moving into law, medicine, and finance. The same dynamic follows: people making consequential decisions with tools they can't interrogate, in domains where "I'll just re-prompt it" isn't a recovery strategy.…