I hadn't considered that. I'm using it almost exclusively to validate logic, kind of like a fuzzer in nature ("What if we need to do this with this logic/someone tries to do that/what am I missing/etc"), or to fill in specifications ("what feature would compliment this/what could be trimmed to achieve MVP more easily/does this spec appear to be missing anything according to this set of requirements"), which requires a lot of review, and using more expensive models like Opus doesn't appear to provide meaningfully better results. After prompting it, I typically have a lot to think about and the terminal goes quiet, or I prompt it on a related matter that will similarly require my eyes and brain for long enough that I won't be able to limit out.

I can see how work involving larger contexts and deeper consideration would lead to exhausting limits a lot faster though, even if you aren't using it like a slot machine.