What is already becoming manifest is that AI dev tools like Cursor or Copilot are very useful in the hands of already senior engineers, who know when to call it on its BS. In my org, the senior devs who are using it basically have an incredibly fast junior dev at their disposal to help them get stuff done faster, which is all well and good. But when the human junior dev uses it in the same way, they implicitly trust the output from the AI junior dev and thus there are no guardrails preventing the blind from leading the blind. We just had a junior on my team send up code for review that while (thankfully) not completely wrong was done in such an over-engineered and complicated way that if it were to have been approved it would've been completely out of line with the best practices and general framework of our codebase. Do this enough times and best case scenario you have a terribly unwieldy, spaghetti codebase and at worst a ton of bugs making it to production.