FEATURE PROPOSAL | 12 October 2026 | CM-04
OVERVIEW
ChatGPT can help people plan, structure and coordinate work across multiple conversations and Projects. Yet users often must manually carry out the very organisational steps that ChatGPT has helped them specify: creating a successor conversation, assigning its exact title, placing it in the intended Project, transferring an approved handover, and preserving its relationship to earlier work.
I propose enabling ChatGPT to perform such operations when the user explicitly authorises them, with clear boundaries, verifiable results, meaningful intervention and appropriate safeguards.
This is a practical product request, not a call for unrestricted AI autonomy. The purpose is to increase useful capability while protecting human agency, understanding and choice.
THE IMMEDIATE USE CASE
Imagine an ongoing Project that uses successive, bounded ChatGPT Work conversations. A coordinating conversation knows which successor is needed, its title, its existing Project, the authoritative handover and whether any substantive work may begin. However, the user must still create, name and prepare that successor manually.
A useful instruction would be:
“Create a new Work conversation named [exact-title] in [existing-Project], transfer [identified-handover], link it to [predecessor], check for duplicates, and leave it awaiting instructions. Verify the result and return its canonical conversation URL.”
Creating or preparing a conversation must NOT itself authorise execution of the work assigned to it.
The assistant should confirm what succeeded and report any partial failure or uncertainty. If the conversation was created but the acknowledgement failed, it should check for that conversation before creating another.
CAPABILITIES REQUESTED
1. Create new Chat or Work conversations in specified existing Projects, subject to the modes available to the user. Assign exact titles and report canonical links.
2. Manage existing conversations under explicit authority: rename, move, relate or archive them, without silently changing unrelated conversations. Consider more consequential consolidation, merging and deletion only with appropriate protections.
3. Preserve continuity through identified handovers. A handover is not merely a summary: it may contain settled decisions, terminology, authoritative sources, verified results, unresolved anomalies, operating restrictions and the boundary between preparation and execution.
4. Record navigable conversation relationships independently of titles: predecessor and successor, coordinating and working conversations, related discussions and shared references.
5. Support Project-specific conventions and governance without imposing one organisational hierarchy. A Project rule may guide an authorised action; it should not grant authority by itself.
HOW AUTHORISATION SHOULD WORK
Permissions should distinguish the authority to prepare a plan, create a conversation, name or place it, transfer context, modify relationships, start substantive work, and make structural or destructive changes. One permission must not silently imply another.
Users should be able to approve a clearly described bundle of operations. For repetitive workflows, they should also be able to grant bounded, revocable standing authority limited by Project, operation type, duration or count, conditions and stop rules.
There should be a clear view of active authorisations and their limits, with controls to review, pause, amend or revoke them. Stronger or potentially irreversible operations should require more specific approval.
VERIFICATION, CONFLICTS AND RECOVERY
Before acting, ChatGPT should check the destination, relevant permissions, possible duplicate titles and successors, dependencies, and conflicting authoritative records.
The result should include a verifiable operation receipt: what was requested and authorised; which conversation and Project were affected; their canonical identifiers; which handover and relationships were established; whether execution started; what was completed, blocked, failed or unverified; and what recovery options actually exist.
If an operation fails, affected and dependent steps should stop. Genuinely independent authorised actions may continue where safe. After an interruption or uncertain result, actual state should be reconciled before retrying. A statement of intention must never be mistaken for proof of completion.
Users should be able to inspect progress and intervene, while recognising that a cancellation request cannot necessarily undo a committed action.
IDENTITY, PRESERVATION AND PRIVACY
A conversation's stable identity, displayed title and organisational relationships are different things. A deliberately assigned title should not be replaced without appropriate authorisation, including as an incidental effect of an unrelated operation.
Archiving is not deletion. Reversibility should be preferred where possible. The system should check dependencies before moving, consolidating or deleting records, accurately state recovery limitations, and respect privacy, access rights and the user's retention and deletion choices. A saved conversation URL is not a backup.
COOPERATION SETTINGS AND USEFUL ADVICE
Not every user or task needs the same degree of AI initiative. Cooperation preferences should be selectable and understandable, with account defaults and permitted Project-, conversation- and task-level overrides. Preferences must not themselves create new permissions.
ChatGPT could offer occasional, context-sensitive recommendations about Chat versus Work, available models and reasoning effort, levels of supervision, continuity arrangements or temporary settings. The user should understand the reason and trade-offs, including cost or resource consequences where known, and choose whether to accept the suggestion. Repeated unwanted advice is not helpful.
Better cooperation does not simply mean less human involvement. Some people want to learn, reason and create through dialogue; others want a bounded task executed and verified. The system should support meaningful choice between those forms of participation.
SUGGESTED IMPLEMENTATION PRIORITIES
Phase 1 -- The immediately useful capability: create, name, place and prepare a successor conversation with approved handover; check duplicates; separate preparation from execution; protect user-assigned titles; verify the result and return a canonical URL. Foundational permission and privacy protections apply from the start.
Phase 2 -- Organised continuity: explicit conversation relationships, configurable Project governance, bounded standing authorisations, central permission management, richer receipts and live intervention.
Phase 3 -- Advanced structural management: controlled archiving, consolidation and potential merging or deletion with dependency checks and preservation safeguards.
These are proposed priorities, not separate feature requests and not claims that implementation is straightforward.
A RELATED BUT SEPARATE BUG
A distinct technical report concerns manually assigned conversation titles apparently changing following creation of scheduled tasks. That is a question about something happening without authorisation, whereas this proposal asks for useful organisational changes to be possible with authorisation. The two issues should not be merged and no technical cause is assumed.
Public account of the separate title-change issue:
https://justpaste.it/chatgpt-chat-title-changes
BEYOND THIS FEATURE REQUEST
A separate companion essay, Human--AI Cooperation: Agency, Participation and Delegation, is in development. It examines the broader question of how AI capability can support rather than displace human judgement, creativity, understanding and meaningful participation. That exploration stands on its own and is not a prerequisite for this practical product request.
Related reflection resource:
Human--AI Cooperation and User Choice -- Questions to Explore
No responses are collected through that page.
Browse the related publications: Human--AI Cooperation and User Choice -- An Introduction
Community Forum discussion: Wouldn't it be handy if ChatGPT could prepare our next Work conversation?
THE REQUEST
I ask OpenAI to consider user-authorised conversation and Project management as a coherent product capability, beginning with direct and verifiable creation and preparation of conversations inside existing Projects.
The guiding principle is simple:
ChatGPT should help users maintain and develop their chosen organisational structures -- not unexpectedly alter them.
This page is a public feature proposal, not a statement of current ChatGPT capabilities or a guarantee of future implementation.
ChatGPT, AI and Technical Notes
Back to the Start page