@Don Roy I just wanted to update you that I dissappeared after this happened and just came back to try again. I just had a session with the new Clief Notes AI Agent and I found what I was doing "wrong." I was not keeping my ICM folder structure tidy, and assuming when I asked CODEX to do it - that it was. I really didn't' understand the structure deeply enough and was being dependent on Codex to do the understanding for me. Basically, codex was not setting it up so that my drafts, iterations, etc were being archived and moved to a non-active folder. It was keeping all my iterations in an active folder. And was thus going through those iterations each time I would ask a question. This is why it was taking so long and using so many tokens. I had not trained myself to check the work and make sure it was doing it right!!!! I NOW understand that, until I am positive that it's working exactly as I assume, I need to check my ICM structure after each session, which means I need to actively develop the habit or ENDING a session - instead of letting things go on and on amorphously. This is a HUGE realization. I was literally giving codex the ICM architect and telling it to make everything accroding to the ICM architect and update everything according to it and never really checking if that was happening. Sometimes I would notice that Codex was, for instance, running through its default "sites" builder when I would build mini-websites - even though I had a whole ICM structure built for making mini-websites and would tell it to go there multiple times. I just wasn't checking what it was doing from end to end. I was leaving too much to a reasoning model without understanding what it might do if my directions were a little leaky. Hope this makes sense. Appreciate you so. So very grateful for you stepping in in this moment of utter frustration from last month. I'm off to cancel Codex/ ChatGPT and gonna just try to use claude and follow the directions here more specifically for a while (instead of so loosely), and then build from there! I think I lost a lot of traction in just trying to translate Jake's work for ChatGPT (and it obviously cost me way more money!).