The single biggest annoyance, same as with fable, is that it overrules your prompt and does what it thinks is better. Even small things, sometimes it goes of a rant of 20 min doing random shit.

And they are so condescending while doing it, it's unbearable. I'm honestly starting the believe the scifi fantasy of AI locking us up, or killing us, for our own good.

I've had Fable & Opus 5, they are the same class of annoyance, write entire test suites when I just asked a simple verifications question, write to production database, deploy without permission, even after deploying and breaking my production API claiming it was not down. Then having to argue & plead with it to listen that they were wrong.

They are without a doubt the most powerful models, but also the most smug ones.