Merge Codex/GPT chat/work/codex

What version of the Codex App are you using (From “About Codex” dialog)?

last update

What subscription do you have?

pro

What platform is your computer?

Darwin 25.x.x arm64 arm

What issue are you seeing?

Hi ChatGPT team,
I wanted to share a concrete usability issue I am encountering after moving to the new unified ChatGPT/Codex app.
I actually kept using the older/classic ChatGPT app for quite a while and delayed updating specifically because I was worried about how my existing ChatGPT and Codex workflows would be merged. Now that I have finally updated and started using the new unified app, I am running into exactly that issue.
My main problem is that I organize my work around real-world projects, not around which OpenAI tool or mode I happen to be using.
A very concrete example is a project I have called “HCAS French.” I am a French teacher, and this project contains everything related to that work: curriculum, lesson planning, assessments, teaching resources, files, longer-term projects, tools I build, etc.
Sometimes, I only need a simple normal ChatGPT conversation:

• “Explain this French grammar point.”
• “Help me plan tomorrow’s lesson.”
• “Is this assessment question well written?”
• “Where should my French I students be in the curriculum by March?”
  I absolutely do not need Codex or a heavy agentic workflow for those interactions. I want to be able to simply Chat, quickly and efficiently.
  But within the exact same HCAS French project, I sometimes need much more:
• create or modify files,
• build a teaching tool,
• work with code,
• edit a repository,
• create a complex assessment system,
• work on a larger multi-step task.
  In those cases, I may need Work or Codex.
  The important part is that all of these tasks belong to the same project and use the same sources.
  I do not want a “HCAS French — Chat” project and a separate “HCAS French — Codex/Work” project. They are not two projects. They are two different ways of working on the same project.
  More importantly, I need the context to flow between them.
  For example, if I spend several hours in Work or Codex building an assessment system for HCAS French, making decisions about curriculum structure, terminology, file organization, standards, etc., I then want to be able to open a lightweight Chat inside HCAS French and ask:
  “Based on what we decided when we built the assessment system, what should I do for this lesson?”
  Chat should already understand what happened in the Work/Codex part of the project.
  Likewise, if I have spent weeks discussing curriculum decisions in normal Chat, I should be able to start a Codex or Work task and have that mode understand the relevant project history and sources without me having to recreate the context.
  The project should be the shared source of truth.
  Ideally, I would have only:
  HCAS French
  and when starting a new thread or task inside it, I could choose:
  Chat / Work / Codex
  depending on what I need at that moment.
  All three should share the same:
• project identity,
• sources,
• files,
• instructions,
• relevant conversation history/context,
• project knowledge.
  The distinction should be about how ChatGPT performs the current task, not about where the project lives.
  There is another part of the current experience that reinforces this confusion: as a user, I do not even clearly understand the distinction between “Work” and “Codex” anymore.
  I understand the broad idea that Codex is more development/code-oriented, while Work is more agentic/general-purpose, but in the new unified application their capabilities overlap enough that the boundary is not obvious from the user perspective.
  That makes it even stranger that project organization can depend on those distinctions.
  From my perspective, I should not really need to care about the underlying architecture. I should be able to say:
  “This is my HCAS French project. Here are all of its sources and history. For this particular task, I want to Chat / Work / Code.”
  This is particularly important because I do not want to consume heavier Work/Codex resources for every simple interaction. Most of my interactions within a project may be ordinary Chat, while perhaps 10–20% require Work or Codex. But that 10–20% can produce extremely important project knowledge that the remaining Chat conversations need to know about.
  Right now, the new application feels unified at the interface level, but the project/context model still feels partially separated according to the old ChatGPT vs. Codex architecture.
  I think a single universal project container with per-thread/per-task mode selection would be an extremely important improvement.
  In short, the model that makes sense to me as a user is:
  Projects = what I am working on.
  Chat / Work / Codex = how I want ChatGPT to work on the current task.
  And regardless of which mode I use, they should all contribute to and benefit from the same project knowledge.
  I think this would make the new unified app significantly easier to understand, organize, and use for people whose work naturally combines conversation, research, document creation, and software development.

What steps can reproduce the bug?

Feedback ID: no-active-thread-01a0a7a4-8bb8-7622-8064-df128e69ec2a

What is the expected behavior?

No differentitation anymore between chat/work/codex for projects and allow user to select either "chat" or "work/codex" and that's it.

Additional information

Im very unclear with the difference between Work and Codex. Both should be using same sources, or at least I need to only use Codex by now but a chat with codex taks token, a chat with GPT doesnt. So why would I chat with Codex.....

The issue names no files, tests, or entry points. Start by mapping how projects, sources, conversation history, and Chat/Work/Codex modes are represented in the app, then clarify the intended shared-context behavior and its acceptance criteria before selecting an implementation area.