Chat mode gone in VSCode extension!

Chat mode was removed in favor of the new "plan mode". This was an intended change, not a bug.

Chat mode was removed in favor of the new "plan mode". This was an intended change, not a bug.

WTF? In which world does "Plan" means the same thing as "Chat"?

Plan mode in the context of AI agents universally means writing down some spec, usually a markdown file with a multistep plan to execute a large coding task. It heavily implies the Agent is pre-instructed to be in a planning mode putting together user received information into a specific bulletpoint plan.

No one in their right mind would use "Plan" mode for a simple one off questions about the code. This is so incredibly confusing and bad UX.

This is just so retarded beyond comprehension. I tried to use Plan mode for quick chat question:

<img width="559" height="762" alt="Image" src="https://github.com/user-attachments/assets/d167bbe8-5704-40c4-8f0f-51b8923bbb67" />

The Codex immediately started to build a plan instead of answering the question:

<img width="588" height="1018" alt="Image" src="https://github.com/user-attachments/assets/b2b9f889-ee96-4f89-a2cc-1ffed3ba7c79" />

And then proceeded to ASK ME questions instead of answering the questions I asked.

This is clearly a mode to put down a spec, not a mode to ask questions without modifying the code.

I refrain from using harsh language but i concur with the problem: Chat mode is what i want by default, agentic when i choose to.

I know that i can place an .md-file to restrict the LLM, but this does not feel like a safe way to do it, because who knows what can cause a breach of that instruction.

A secured way to chat and approve code changes one by one is what fits my workflow, the current solution is not what i want to use.

I would love to hear thoughts from other people about this and the reasoning behind the change.

EDIT:
Short note: It seems that the following changes to the config.toml enable approval requests for all actions:

sandbox="read-only"
ask-for-approval="on-request"

I refrain from using harsh language but i concur with the problem: Chat mode is what i want by default, agentic when i choose to. I know that i can place an .md-file to restrict the LLM, but this does not feel like a safe way to do it, because who knows what can cause a breach of that instruction. A secured way to chat and approve code changes one by one is what fits my workflow, the current solution is not what i want to use. I would love to hear thoughts from other people about this and the reasoning behind the change. EDIT: Short note: It seems that the following changes to the config.toml enable approval requests for all actions: `` sandbox="read-only" ask-for-approval="on-request" ``

Unfortunately sandbox doesn't work, because the agent is not aware it's in ask mode. So it will waste a LOT of time and reasoning budget on trying to apply the patch, then realize it doesn't have the tool available, and only then fall back to putting the answer in chat.

In practice, it means you will be waiting a lot for every answer and burning a lot more reasoning limit quota than you need to.

No chat mode pretty much make codex unusable. As you definitely don't want AI agent to wreck havoc on the project when you just ask it about something.

Yes. This has been a big pain in the recent update.. Why would someone remove something that works well for the users? Are they not conscious of this.

<img width="621" height="171" alt="Image" src="https://github.com/user-attachments/assets/1aeebe3c-f34a-4eae-85d7-faf429e0d7a6" />

is it just me or I made some mistake on configuring it because I saw no mode switch.

This is incredibly baffling as someone that just wants to be able to use this essentially as a pair programming buddy there is no way to use it like that without config changes.

Not everyone wants to let an AI go loose on their project please change this...