Desktop UX regressions in the unified ChatGPT/Codex app

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #31847
• #31845

Powered by Codex Action

Additional desktop UX issues:

1. New chat icon is confusing

The pencil icon in the upper-left corner is commonly understood as “edit” in many applications. In this interface, it appears to mean “New chat.”

This is confusing because the icon does not clearly communicate that it will start a new conversation.

Suggested improvement:
Use a more standard “plus” icon for New chat, or add a clearer label next to the button.

1. Floating window cannot be resized

The floating conversation window appears to have a fixed size or limited resizing behavior. This makes it harder to work with longer conversations, images, and generated content.

Expected behavior:
The window should be freely resizable like a normal macOS application window.

Suggested improvement:
Allow floating conversation windows to be resized, maximized, or opened in the main application window.

Unexpected Codex usage / credits shown after attaching chat to task

After using Attach to task, the conversation moved into the Codex task interface. The sidebar then showed duplicate task entries, and the app displayed a usage warning: “6% usage remaining” with options to Add credits or Upgrade.

This is confusing because I was not intentionally running a Codex coding task. I was only using the app to discuss feedback about the desktop interface.

Expected behavior:
The app should clearly warn before converting a normal chat into a Codex task if that may consume Codex usage or credits. It should also distinguish normal ChatGPT chat usage from Codex task usage more clearly.

Suggested improvement:
Before attaching a chat to a task, explain whether continuing in that task will use Codex credits. Also provide a clear way to return the conversation to normal chat mode.

Attach to task unexpectedly creates task entries, but nested view is useful

After using Attach to task, the conversation moved from a floating window into the main application workspace. This nested view is preferable to the floating window because it keeps the conversation inside the app.

However, the behavior was not clear. The app did not explain whether the chat was being converted into a task, attached to an existing task, duplicated, or moved.

Two task entries with the same generated title appeared in the sidebar:
“Fix pencil edit and resize”

Expected behavior:
The app should clearly explain what Attach to task does before creating a task. It should avoid duplicate task entries with the same name.

Suggested improvement:
Provide an option to open chats in the main application window without converting them into Codex tasks or consuming Codex usage. The nested-in-app conversation view is useful, but it should be available as a normal chat display option too.

Codex usage decreased from 8% remaining to 6% remaining over roughly 30 minutes without intentional coding activity.

During that time, I was not asking Codex to edit code, run commands, or work on a repository. The only recent action was attaching a normal chat to a task, after which duplicate task entries appeared.

Expected behavior:
The app should show exactly which task or action consumed usage. Normal chat or viewing an attached conversation should not silently consume Codex credits without a clear warning.

Task status and usage transparency
The task list displays elapsed times (for example, 1m, 9m), but it is not clear whether a task is actively running or simply showing elapsed time.
Because Codex usage is limited, the application should clearly indicate:
Whether a task is currently consuming usage.
When a task has stopped.
When a task is waiting for user input.
Whether background tasks continue consuming usage while the user is reading or away from the computer.
Without this information, users cannot determine whether tasks are idle or continuing to consume their available usage.
Suggested improvements:
Add a clear Running, Paused, Waiting, or Completed status.
Show when usage is actively being consumed.
Allow users to stop or pause tasks directly from the task list.
Display an estimate of usage consumed by each task.

Codex usage decreased from 8% remaining to 6% remaining over roughly 30 minutes without intentional coding activity.

During that time, I was not asking Codex to edit code, run commands, or work on a repository. The only recent action was attaching a normal chat to a task, after which duplicate task entries appeared.

Expected behavior:
The app should show exactly which task or action consumed usage. Normal chat or viewing an attached conversation should not silently consume Codex credits without a clear warning.

Task management lacks lifecycle controls. The UI provides a way to attach conversations to tasks, but there is no obvious way to stop, pause, detach, or end a task. If tasks can continue processing or consuming usage, users need clear controls and status indicators to manage them.

Another regression: the right sidebar is now way too narrow by default - I used to rely a lot on double clicking the divider to snap back to 40% or whatever, and now it snaps back to like 10%.

Desktop navigation prioritizes Codex over Chat
The desktop application is heavily centered around Codex tasks, making conversational ChatGPT workflows less prominent.
Observed behavior:
The primary navigation emphasizes Tasks and Codex workflows.
Chat conversations feel secondary and open in floating windows rather than within the main application.
Conversation history is separated from the primary workspace.
The only items that can currently be pinned in the sidebar appear to be Codex tasks and projects.
ChatGPT conversations and ChatGPT Projects are not given equal prominence.
Tasks can be archived, but there is no obvious option to delete them.
Suggested improvements
Allow users to choose a Chat-first or Codex-first layout.
Give Chat and Codex equal visibility in the main navigation.
Allow ChatGPT conversations and ChatGPT Projects to be pinned.
Provide both Archive and Delete options for tasks.

"Attach to task" can be activated unintentionally
The Attach to task action is easy to activate during a normal ChatGPT conversation.
Once attached:
There is no confirmation explaining what will happen.
There is no obvious way to detach the conversation.
The task displays an increasing elapsed time, but the interface does not explain what it represents.
It is not clear whether the task is active, idle, completed, or consuming usage.
This creates uncertainty about whether a normal conversation has become an active Codex task.
Suggested improvements
Confirm before attaching a conversation to a task.
Clearly explain what attaching a conversation does.
Provide an obvious Detach from task option.
Display clear task states such as Running, Waiting, Paused, and Completed.
Clearly indicate when a task is actively consuming usage and when it is not.

Work vs. Codex modes are not clearly differentiated
The application provides two primary modes:
Work
Codex
However, switching between them does not noticeably change the interface, navigation, or organization of the application.
For a user, it is not clear what changes when selecting one mode over the other.
Suggested improvements
If these are intended to be separate workflows, the interface should clearly reflect that.
For example:
Work
Designed for general ChatGPT usage:
Chat conversations
Chat history
ChatGPT Projects
Pinned conversations
Images
Documents
Scheduled reminders
Writing and research
Codex
Designed for development work:
Coding tasks
Repositories
Worktrees
Git
Local environments
Terminal
Task history
Coding sessions
This would make the purpose of each mode immediately obvious.
The current interface presents nearly the same navigation for both modes, making the distinction difficult to understand.

Floating conversation window displays incorrect resize behavior
The floating conversation window displays the standard macOS resize cursor when hovering over the window edges and corners.
However, the window cannot actually be resized in those directions.
Observed behavior:
The left and right edges display the horizontal resize cursor, but the window width cannot be changed.
The corners also display resize cursors, suggesting diagonal resizing is supported.
Dragging from the corners does not resize the window.
The title bar allows the window to be moved, but it does not allow resizing.
The resize cursors indicate that horizontal and corner resizing are supported, but dragging has no effect.
This creates the expectation that the window is resizable when it is not.
Suggested improvements
Only display resize cursors for directions that actually support resizing.
Or allow the floating conversation window to be resized from its edges and corners using standard macOS behavior.

After generating an .xlsx file, the web app crashed. The conversation continued in the desktop app, but the generated file attachment is only visible in the web version of the conversation and not in the desktop app.

The conversation itself remains synchronized, but generated downloadable files do not appear consistently across clients after the crash.

Expected behavior:
Generated files should synchronize across both the web and desktop applications regardless of which client created them or whether one client crashed.

Chat window does not remain scrolled to the latest response.

After I submit a prompt and the response finishes, the conversation frequently remains positioned higher in the thread instead of moving to the end of the newest response.

The down-arrow indicator appears, and I have to click it or manually scroll to the bottom every time.

Expected behavior:

• After submitting a message, the view should follow the active response.
• When generation finishes, the latest response and message box should remain visible.
• The user should not have to manually scroll after every exchange.

Current behavior:

• The view stops partway through the conversation.
• New response content continues below the visible area.
• The conversation does not return to the bottom automatically.
• This happens repeatedly in the floating Chat window.

This makes longer Codex and ChatGPT workflows frustrating because every response requires an additional scroll action.