Honestly, this is why I’m still all-in on Codex. Things won’t always be perfect, but the team listens, investigates, explains what happened, and ships fixes fast. The full reset is a pretty nice bonus too.

I switched to Codex because it started feeling less like an AI coding assistant and more like a place where I can actually build things. I use it for pretty much everything now — building apps, fixing bugs, working with servers, GitHub, deployments, and sometimes just taking a random idea and seeing how far Codex can get on its own. The biggest thing for me is that I don’t want to spend all day telling an AI exactly what file to edit or what command to run. I want to explain what I want, give it the context, and let it work. That’s probably what I like most about Codex: the agentic side of it. When it understands the project well and keeps going without needing me every few minutes, it feels incredibly powerful. I also really like the desktop app experience. I’ve gradually found myself spending more time in Codex and less time jumping between different tools. The biggest improvement I’d love to see is making that autonomy even more reliable. Better long-running tasks, better context management, fewer unnecessary interruptions, and better recovery when something goes wrong. I want to be able to give Codex a fairly big task and trust that it’ll keep working, make sensible decisions, test its own work, and come back when it’s actually done. Also, small UX things matter a lot when you use it every day. For example, voice input in the built-in terminal currently submits my text when it gets long enough to wrap, which makes it basically unusable for longer prompts. For me, the direction is pretty simple: I want Codex to become the place where I can go from “I have an idea” to “this is built and working” with as little babysitting as possible. That’s the reason I switched, and honestly that’s what keeps me using it. @thsottiaux