Just today, I asked Sonnet 3.7 to use the app's accent color, which can be found in the @global.css, to apply it to some element. It went on a massive tangent, scanning all sorts of documents, etc. I then asked Sonnet 3.5 for the same thing, and it simply took the accent color and applied it. No tangent, no grep, ...
It does things I didn’t ask, while deleting random things I just asked to add in the previous prompt, it’s a mess.
However it does one thing very well: it can make me angry very quickly, like a real human.
Definitely can be very annoying if you do just want it to execute on a set of instructions.
To me it's not a good deal that I pay for something that makes up the things it wants, and completely disregards what I asked. It's like the recipe for the worst SaaS you ever used.
Thankfully, switching to Sonnet 3.7 in 'Ask' mode has been essentially fine and is comparable to how Sonnet 3.5 performed for me.