Just the other day I wasted 20 minutes tracking down a problem before caving and pasting some code into Claude, running a massive machine learning model trained by the foremost researchers in the AI industry, all so it could tell me I forgot semicolons in some CSS import lines.
I thought we were done with such goofiness in the days of auto-linters and syntax highlighting, but that’s why it was so tricky! The cause is always the last thing you think to check, after all…
I get good mileage out of this as well. Sometimes you know that a bug is dumb and you still don't see it right away. LLMs are good at dumb bugs, and I would much prefer to spare my effort for the not-dumb ones.
That, and a head start on scripts, or well-defined functions which are just fiddly enough that I'd have to think about them, are most of the work I get out of our new chatbot junior developers. I still find them broadly useless for deep work, but no one said they have to be.
I thought we were done with such goofiness in the days of auto-linters and syntax highlighting, but that’s why it was so tricky! The cause is always the last thing you think to check, after all…