OpenAI's transition from Codex to ChatGPT is a workflow
The real problem isn't just the UI change; it's the lack of context separation. When you download what is marketed as the "Codex" replacement, you just get the full ChatGPT suite. This means your entire work-related prompt history is now tangled up with your personal life (because heaven forbid I see a grocery list while I'm debugging a production kernel panic).
OpenAI's documentation makes it sound like there's some "Codex mode" available within the ChatGPT interface, but there is absolutely zero visual indicator or UI toggle to enable it. It’s all smoke and mirrors. I spent way too much time digging through their GitHub releases trying to find a standalone macOS binary that actually functions like the old version, but instead, I just found a chaotic mess of CLI tools and app servers.
If you are trying to reconstruct a specific prompt structure that worked in the old environment, you might be able to force the behavior by being extremely explicit with your system instructions, though it's a far cry from having a dedicated workspace.
For those trying to replicate the old logic in the current ChatGPT environment, you basically have to wrap your intent into a strict system prompt to stop the model from drifting into "assistant mode" and keep it in "coder mode."
You are a specialized Codex-mode engine. Your sole purpose is to act as a high-precision code completion and refactoring tool. Ignore all conversational filler. Do not engage in small talk. When provided with code snippets, analyze them for logic errors, efficiency, and adherence to best practices. Provide only the corrected code or concise technical explanations.It’s a frustrating shift from a streamlined tool to a bloated general-purpose chatbot. We're losing the "tool" feel in favor of a "companion" feel, which is great for consumers but a headache for engineers who just want to get their work done without the cognitive load of a cluttered chat history.