using codex and gemini flash 3.8?!

Okay, recently I've been having some big, big problems with AI. It's just been driving me crazy, and this is definitely not a complaint post. What I want to talk about is (some of) my old workflow problems, and the solutions to them. --- Old - generalising, this would involve two major steps: 1. plan as i normally would -> 2. implement incrementally The implement part was becoming the problem, it felt like slowing banging my head against a brick wall trying to make sense of verbose output, questioning every output (which on some level i think is needed ofc) and getting exhausted just trying to understand what was happening. Most of this was with Claude models on the Max x20 plan, which i've used and loved up until the last 2 months... as probably most of you guys know, the output tokens, the way the model responds to us, is not that great. It is often very verbose and very dense in the information. You have to spend a lot of time figuring out what's relevant and what's irrelevant, and then imagine context switching between the rest. This was becoming exhausting. --- New - i started by remembering i can agency, so w hat I did was cancel my Anthropic sub , and I worked with a new workflow. I've switched to Codex, and I run this workflow. I use, first of all, Luna or a Terra thread, and start the session there. I give it instructions to use Astra for reviews, for clarification, for anything that is the slightest concern, but not to give it the full context. Just give it the context that's relevant to that scenario. Ask for feedback. The goal here is to give Astra the key decisions, but to take away all of the context burn or token burn, and not give a ton of information to Astra. --- The surprise model that's working in professional and personal projects I'm using Google Flash 3.8?! first of all hold your judgement until you try. it is so, soooo fast. I think the inference on this is just faster than any of the other models I'm using. It seems to me I'm using anti-gravity, and it's just sped up my workflow so much. That is taking a lot of cognitive effort as well. --- Guys, hope this helped. I'd love to hear if you guys have different workflows or if you could improve it in a certain way.