That Opus 4.5 hype is real but the day-to-day variance point matters more than people admit. Had a project last week where it nailed complex refactoring on Monday then struggled with basic state management on Wednesday. The trend toward teams moving away from AI coding tools (that Hunter tweet) is fascinating becuase it suggests we're hitting the review bottleneck faster than the generation one.
That Opus 4.5 hype is real but the day-to-day variance point matters more than people admit. Had a project last week where it nailed complex refactoring on Monday then struggled with basic state management on Wednesday. The trend toward teams moving away from AI coding tools (that Hunter tweet) is fascinating becuase it suggests we're hitting the review bottleneck faster than the generation one.
Yes, good point that it's not stable!
There's even a tweet about potential degradation this morning from Claude team: https://x.com/trq212/status/2001541565685301248