I agree, I also noticed visible drop in quality, a week before I was able to ship two apps and when I was working on the third (which is almost as much big as the first) it started acting weird, I even started a new session thinking it was because of hallucination... It did give me the output but burnt through my tokens earlier than I expected. It was even missing and getting confused between branches... All in all doing more mistakes... One point I would like to bring to your attention is, it has started using background agents a lot lately (maybe because I prompted it to use Sonnet models for scripting and Opus will supervise, but still it was better in the earliest week.)