Hey builders! 🚀 Quick question for anyone currently architecting or messing around with an AI OS setup. I’m weighing my options for the core engine and looking closely at OpenAI Codex. Before I commit fully to it, I wanted to pick the brains of this group: 1. Is Codex a solid foundation for an AI OS? How well does it handle the background automation, system-level task execution, and sandbox management compared to other models out there right now? 2. What’s your actual build experience? If you’ve built or prototyped an AI OS using Codex, what did the reality look like? Any major roadblocks with context windows, agentic drift, or state management? 3. How generous is the usage on the high-tier plans? Specifically, if I'm looking at the $200 Pro subscription with 20x usage limits, does it actually hold up under heavy, multi-step agentic loops, or do you still find yourself hitting a wall during intense dev sessions? 4. Would love to hear your raw feedback, architecture tips, or if you ended up switching to a pay-as-you-go API model (like Claude Code) for better reasoning and fewer caps. Also, a huge shoutout and thanks to Nate for putting this community together. The insights dropping in here lately have been gold. Drop your thoughts below! 👇

A year ago, I was still convincing people that AI assistants could be trusted with real work. This week, I watched a security team approve Claude for company-wide use in under an hour. The reason? The new Compliance API integrations mean they can govern it the same way they govern everything else. That's the quiet shift happening right now. It's not about flashier models (though Opus 4.7 is genuinely impressive). It's about Claude becoming something that fits into how organizations already work — governance, audit trails, self-hosted sandboxes for agents, identity controls. The May 2026 updates aren't trying to wow anyone. They're trying to remove the last reasons not to use it ,And honestly? That's the more interesting story. What's been the unlock for your team?

I keep hitting my Claude Code token limits and there seems to be so many different ways to reduce token usage. What have you found to work the best?

I built an emaildesigner and I managed to make it into a command, that is new and exciting. I changed the colorings to get the design more layered and pop more. i still want to work on the sales scripts, but I will come back to that. I learnt to move on. When I have learnt more, which I do every step here, it is easier to get the output I want.

Few days ago I started to practice and code an aios type project with the help of claude code. I have learnt a lot and hope to reach level 3 to unlock Nate's tutorial for it to get more knowledge of what I'm lacking right now. Good luck on your projects guys!😄