Jesus christ why?!
I just asked codex a simple question, it and immediately proceeded to edit my code. This is just such an insanely giant downgrade! Worse yet, the UNDO button in the Codex extension UI doesn’t even work, so I could not even revert the changes codex made, and had to go file by file reverting them manually.
This is crazy! Why would you remove it! Just let me ask questions about the code without Codex immediately jumping in to mess with it.
Are you seriously implying constantly editing my Agents.md or having to constantly tell the agent some key phrase is a viable alternative to a simple “chat/agent” switch button?
Your concern is that the AI eagerly emitted a tool call.
The “developer” message is the place to tune that behavior, or prohibit it with no latitude for judgements.
The only “constantly need to” in my case that the AI is shown understanding, that the user will communicate clearly when to start patching the discussed improvements.
As for “why did?”
Consider that OpenAI is trying to push people into a product that will create subscription upgrades and dependence on a product that will cost additional prepaid use in ChatGPT. Making the promo preview free for onboarding.
A choice could have been made, “no, we don’t want to simply give free users free buddy-buddy chats with an expensive coding model.”
@_j ‘s solution is a good short term fix that works for me.
@rawalanche 100% agree it’s wild they would remove a core feature - and without any warning or documentation. I kvetched about it here. @vb made a good point that in the CLI, you can still adjust the “approvals” setting with slash commands /permissions .
