Claude tells me No more often than Codex

I've been a heavy Codex user for a while and recently switched to Claude for my work which involves lots of different tasks from knowledge work, reports, coding, data science etc. One strength I've noticed is that Claude, well at least Opus, does a better job creating something one shot that is nearly perfect for my needs, both aesthetically but also the insight. But one of the weakness is that sometimes I ask it do something and it simply says I can't and when I ask it for help solving it, it can't figure it out. For example, one might be editing a Google Doc without having the google doc connector set up. It simply wouldn't edit the doc and suggested I copy/paste. Or recently I wanted it to take over computer use to help with my google cloud account settings and it said no because it was too personal and wouldn't even take a screenshot of my safari browser so it could see where I was stuck. I think GPT is still more scrappy and helpful than Claude models, although Opus 5.5 on high wins out on quality.