Feature Proposal: One Conversation, Multiple Modes, Uninterrupted Work
I’m proposing an improvement to ChatGPT’s conversation architecture that would allow users to switch between Chat and Work, communicate during active execution, and queue follow-up instructions without creating new conversations or unnecessarily interrupting ongoing work.
I use ChatGPT extensively for interconnected creative, academic, and business projects. One of the biggest workflow challenges is that human thought doesn’t necessarily follow the same sequence as an AI execution task.
While ChatGPT Work is researching, editing files, or implementing a project, I often identify additional requirements, questions, corrections, or future tasks.
I don’t necessarily want those thoughts to interrupt the current assignment, but I also don’t want to lose them.
Opening another chat within the same Project may preserve the active Work session, but it fragments the project’s history. Decisions, instructions, and deliverables become scattered across conversations.
I believe conversations should function as persistent collaborative workspaces, while Chat and Work should be selectable experiences within them.
Three Connected Feature Requests
1. Bidirectional Chat/Work Switching
Allow users to switch an existing conversation from Chat to Work and back again.
For example, I should be able to brainstorm an idea in Chat, switch to Work to implement it, and then switch back to Chat to discuss the results without starting another conversation.
Model selection and reasoning effort should remain separate from the Chat/Work experience selector.
2. Noninterruptive Messaging
Allow new messages to be received while an active Work assignment continues.
A new message shouldn’t automatically cancel, restart, or redirect work.
Instead, users could ask conversational questions or provide additional context while Work maintains its execution state.
Relevant instructions could be integrated at appropriate checkpoints. If completed work requires revision, the assistant could identify the affected steps rather than restarting the entire assignment.
Urgent interventions should still be possible.
3. Intelligent Follow-Up Queue Within the Same Chat
Introduce optional message-handling controls:
-
Send Now: Address an urgent instruction immediately.
-
Add to Current Task: Incorporate new context at an appropriate checkpoint.
-
Queue as Next Step: Complete the request after the current task.
-
Queue as Separate Task: Create a separate execution objective within the same conversation.
-
Save as Idea: Preserve the thought without initiating work.
These options could live in a compact menu rather than adding five prominent buttons to the composer.
Critically, queued messages should remain in the original conversation.
The queue would be an organizational view of messages already in the chat, not a separate conversation or workspace.
Messages could display statuses such as Queued, In Progress, Completed, or Awaiting Approval, with links to their eventual results.
Concrete Example
Imagine I’m updating a business website through ChatGPT Work.
Step 1: Brainstorm in Chat
I discuss the site’s navigation, artist biographies, and page structure with ChatGPT.
Step 2: Switch to Work
Once we agree on a direction, I switch to Work and ask it to implement the approved changes.
Step 3: Ask a Question Without Interrupting
While Work is updating the website, I realize I want to explore whether individual artist EPKs would be better than a collective media kit.
I switch to Chat and ask the question.
The existing Work assignment continues independently, where supported.
Step 4: Queue a Follow-Up
After our discussion, I decide to create individual EPKs once the website is finished.
I submit that instruction as a queued next step.
The instruction remains visible in the same conversation, with its dependency on the completed website.
Step 5: Add Relevant Context
I remember that all artist biographies should use consistent editorial formatting.
I add that instruction to the current task.
ChatGPT incorporates it at an appropriate checkpoint without unnecessarily restarting the entire assignment.
Step 6: Review and Continue
When the website assignment finishes, its results appear in the same conversation.
I switch to Chat to review them, then back to Work if revisions are needed.
The queued EPK assignment can begin once its prerequisites are satisfied and execution is authorized.
The entire project remains organized in one continuous conversation.
What This Requires Architecturally
I recognize that Chat and Work may rely on different execution environments, tools, models, permissions, and usage limits.
I’m not proposing that those boundaries be removed.
Instead, I’d suggest separating three concepts:
Conversation: The persistent history of instructions, discussion, decisions, and deliverables.
Interaction Experience: Whether the next message requests a conversational response or a Work assignment.
Execution Task: The independently managed process responsible for completing an assignment.
This separation could allow a Work task to retain its state while the user continues having conversational interactions within the same authorized context.
It would also create opportunities for dependency tracking, task-aware navigation, conflict detection, and intelligent handoffs.
Importantly, users should retain clear control over execution, approvals, resource access, and usage.
Why This Matters
For complex projects, the current need to divide related work across multiple conversations can make the experience unnecessarily difficult to manage.
For neurodivergent users, nonlinear thinkers, and people managing multiple responsibilities, immediately capturing an idea without losing their place can also be an important accessibility consideration.
The goal is not unlimited autonomous execution. It’s better continuity, clearer organization, and more natural collaboration.
Suggested Development Approach
Phase 1: Allow Chat/Work switching within existing conversations.
Phase 2: Support noninterruptive message intake while Work is active.
Phase 3: Introduce an integrated follow-up queue with visible statuses and execution controls.
Phase 4: Add intelligent task coordination, dependencies, and conflict detection.
I’d be interested to hear whether others have encountered similar challenges, particularly when managing long-running Work assignments or complex ChatGPT Projects.
The underlying principle is simple: Users should be able to keep thinking while ChatGPT keeps working, without either process unnecessarily interrupting the other.