The Overly Confident Paste
At one point copy and paste was a technological marvel. Eventually it became a development pattern. Blindly copying snippets from around the web into text editors or riskier, directly into your terminal.
Today the workflow has shifted from copying and pasting from StackOverflow to copying and pasting from the output of an LLM. Potentially safer, but also, the robots lie on the regular.
This all became very evident to me last summer as I was noticing an uptick in support requests referencing made up API endpoints and various other inaccuracies. The theme that infuriated me the most was the chest beating confidence that what they were doing was in fact, correct.
To make matters worse, sometimes the thing being pasted is actually technically correct. Case in point, URLs.
You see, URLs aren't new, and their spec is pretty well defined. Until you realize there are multiple (somewhat conflicting) specs. Then you further realize that JavaScript opted to use a permissive spec (WHATWG URL Standard) that allows URLs to contain Unicode-aware identifiers.
What's this mean? It means if somebody copies some output from a robot with an
emoji affinity (as most do) the "URL" could very well defeat validation that
utilizes new URL(...) will say something like this https://example.com/😀 is
just fine.
Technically it is, as the smiley gets normalized and encoded under the hood. The problem gets exacerbated if you're pushing this somewhat tainted content to another system that perhaps is validating against the old school ASCII-based RFC.
Multiple standards combined with confidently pasted AI slop ends up resulting in additional engineering burden. Said burden will probably be solved by asking a different robot how to fix it, and the cycle of overly confident pasting will continue.