They're misleading to consumers in general. I used copilot to write one suite of unit tests for a react form last week and it took ~7 hours of iterations to get functioning tests. Compared to writing them by hand in 4.5 it feels like vibe coding is pure hype.
Nobody is basing their AI predictions on Copilot though. That is the issue with all AI skeptics. Claude Code is absurdly good at editing a software project. Copilot like per month fixed cost tools have a lower ceiling since they are cost constrained.
It might also be that part of the job of a programmer is to translate English (or you preferred human language) to a finished product. I have watched management talk to programmers about what they want. I sometimes wonder how things actually get done.