[Desktop] Suggested prompts can silently switch Chat to Work and consume Work/Codex usage

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

Exact desktop build not captured. Current ChatGPT desktop app as of September 18, 2026.

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Windows desktop app.

What issue are you seeing?

While using normal Chat mode, the ChatGPT desktop app displays suggested prompts/actions that look like ordinary follow-up suggestions.

Some of these suggestions, when clicked, immediately start Work instead of continuing in Chat.

There is no clear indication before clicking that:

• the interaction will switch from Chat to Work;
• Work will be started;
• the request may consume the user's limited Work/Codex usage allowance.

This makes it possible to consume Work/Codex usage unintentionally by clicking what appears to be a normal suggested follow-up.

The problem is specifically the combination of:

1. a suggestion presented inside ordinary Chat UI;
2. no sufficiently prominent Work/usage indication before activation;
3. immediate execution in Work after one click.

This is not the same as manually selecting Work or reopening an existing Work conversation.

What steps can reproduce the bug?

1. Open the ChatGPT desktop app.
2. Start or open a conversation in Chat mode.
3. Encounter an app-generated suggested follow-up prompt/action.
4. Click a suggestion that is backed by Work.
5. Observe that the request starts in Work rather than remaining in normal Chat.
6. No explicit confirmation is shown before the mode change and execution.

What is the expected behavior?

A suggested action should not silently change execution mode when that mode uses a different or more limited usage pool.

At minimum, one of the following should apply:

• Work-backed suggestions are clearly labeled, for example "Run with Work".
• The app asks for confirmation before switching from Chat to Work.
• Suggested prompts respect the currently selected Chat/Work mode unless the user explicitly changes it.
• A setting exists to disable automatic Work invocation and/or Work-backed suggestions entirely.

Preferred invariant:

Never switch from Chat to Work and begin execution without an explicit user action acknowledging the switch.

Additional information

The impact is not merely cosmetic.

Work and Codex usage are limited resources for Plus users. An unintended Work invocation can therefore consume capacity the user intended to reserve for Codex or other agentic tasks.

The UI should make a usage-affecting execution-mode transition explicit before starting the task.

I searched the existing openai/codex issues for suggested prompts switching to Work and did not find an issue describing this exact behavior.

Start by reproducing the behavior in the Windows desktop app using the reported Chat-mode steps, then trace the suggested-prompt handling and the Chat-to-Work transition. Done means a Work-backed suggestion cannot silently switch modes and consume usage without a clear label, confirmation, or explicit mode change.