Like almost everyone I know, I didn’t enjoy working with Claude Opus 5. It was hard to pinpoint exactly what was wrong, but a big part of it was that it would overthink and refuse to be questioned. (It also made me feel stupid, but that might just be a me thing.) The direct impact I noticed in my work was that it would decide on something and stick to it, even when it wasn’t the best or most efficient solution to my problem. Turns out Drew Breunig coined a term for this:
Context Inertia is when the model commits to a hypothesis, subtask, or goal in its context and keeps pursuing it after the evidence says to stop or change course.