Replying to @kenbun
I'm on the $200 plan (which technically gives $5000 of tokens if pushed to the limit all month)
I have a tough time recently pushing my limits on token usage. A month or two ago I was hitting limits much faster, but i think the AI race between Codex vs. CC has forced anthropic to be more generous on their limits
However, my sessions run long, even with Opus 4.8/fable (I mean they are more thorough...). For a decent feature build it could be a solid ~30 minutes. My code base in monolithic though and has about a years worth of LLM gen code on it, so it's definitely not a small project by any means.
When you say a feature takes 30 mins, is it because it’s full stack? Or is it because of how you prompt it e.g. several stages with info on how it should validate each?
Good question...a bit of both. Mostly because it's full stack.
However, I love telling CC to run a GAN loop to produce a feature which basically consists of having it go through adversarial loops (i.e. review + challenge + rework design). And that significantly lengthens the process.
Highly recommend it. Here's an article by Anthropic where they talk more about using it to create long running/productive processes
anthropic.com/engineering/harness-design-long-r…
