> So you read every line. And reading code you didn't write, that was generated by a system that doesn't understand your codebase's history or your team's conventions, is exhausting work.
I’ve noticed this strongly on the database side of things. Your average dev’s understanding of SQL is unfortunately shaky at best (which I find baffling; you can learn 95% of what you need in an afternoon, and probably get by from referencing documentation for the rest), and AI usage has made this 10x worse.
It honestly feels unreasonable and unfair to me. By requesting my validation of your planned schema or query that an AI generated, you’re tacitly admitting that a. You know it’s likely that it has problems b. You don’t understand what it’s written, but you’re requesting a review anyway. This is outsourcing the cognitive load that you should be bearing as a normal part of designing software.
What makes it even worse is MySQL, because LLMs seem to consistently think that it can do things that it can’t (or is at least highly unlikely to choose to), like using multiple indices for a single table access. Also, when pushed on issues like this, I’ve seen them make even more serious errors, like suggesting a large composite index which it claimed could be used for both the left-most prefix and right-most prefix. That’s not how a B{-,+}tree works, my dude, and of all things, I would think AI would have rock-solid understanding of DS&A.
I’ve noticed this strongly on the database side of things. Your average dev’s understanding of SQL is unfortunately shaky at best (which I find baffling; you can learn 95% of what you need in an afternoon, and probably get by from referencing documentation for the rest), and AI usage has made this 10x worse.
It honestly feels unreasonable and unfair to me. By requesting my validation of your planned schema or query that an AI generated, you’re tacitly admitting that a. You know it’s likely that it has problems b. You don’t understand what it’s written, but you’re requesting a review anyway. This is outsourcing the cognitive load that you should be bearing as a normal part of designing software.
What makes it even worse is MySQL, because LLMs seem to consistently think that it can do things that it can’t (or is at least highly unlikely to choose to), like using multiple indices for a single table access. Also, when pushed on issues like this, I’ve seen them make even more serious errors, like suggesting a large composite index which it claimed could be used for both the left-most prefix and right-most prefix. That’s not how a B{-,+}tree works, my dude, and of all things, I would think AI would have rock-solid understanding of DS&A.