Claire found Opus 4.8 impressive for greenfield prototypes and one-shot feature implementation, autonomously coding a working prototype in about 20 minutes. However, it consistently struggled with the 'last 10%' of tasks, including edge cases, finishing touches, and bug hunting, where it often hallucinated by fabricating information. On existing codebases, it required multiple cycles to fix bugs and lacked ambition in creative coding tasks like building a game. In business strategy tests, Opus 4.7 outperformed 4.8 by using data contextually and providing grounded insights, while 4.8 over-rotated on small data points and produced hand-wavy roadmaps. Claire recommends Opus 4.8 for fast, one-shot tasks and new projects but advises caution with existing codebases and strategic work. New features like dynamic workflows and effort control are also highlighted.