How are you guys burning so much of your usage?

I've been working with Astra (xHigh) for the past 2-3 hours, and I've only dropped from 100% --> 99%. This is slow mode, so I'm not paying for 1.5x increased usage. BUT, I do have context set to 1M with compaction to kick in at 900k. This does penalize me 1.5x increased usage. This is on a decently sized code base. I have a pretty intimate understanding of the code, and I'm essentially just asking it to make incremental updates that are highly scoped. If there's a new feature I want, I ask for a mock up so it doesn't have to touch the repo or perform any scans, and then I sit down and think about the easiest way to integrate it into the core app. This keeps input tokens to an absolute minimum. I sort of get the sense that people are just asking it to do extremely large refactors or implementations with ambiguous or underspecified bounds. If you add on Ultracode mode, where you might have 3-4 sub agents all consuming usage, while also doing fast mode, while also having 1M context, then yeah I could definitely see how you might use up 100% of your weekly usage in the span of a few hours.