Feature Request: Native Handoffs Between Chat, Work and Codex Inside the Same Project
I would like to suggest a workflow improvement for ChatGPT Projects, especially for users who use Chat as a project orchestrator and Codex as the implementation agent.
My current workflow is very simple:
- I use Chat as the “project orchestrator”.
- Chat analyzes the problem, reasons about the solution and prepares a technical mission.
- I manually copy the prompt into Codex.
- Codex implements the mission, runs tests and produces a report.
- I manually copy the Codex report back into Chat.
- Chat analyzes the report, validates the result, identifies problems if necessary and prepares the next mission.
This works very well, but the user becomes the manual communication layer between the different agents.
With Work, the same issue becomes even more visible.
For example, a future workflow could naturally be:
Chat → Work → Chat → Codex → Chat
Chat would remain the main project orchestrator.
If deeper research, document analysis or a long multi-step task is required, Chat could delegate it to Work.
Work would complete the mission and return its result directly to the originating Chat.
Chat would then review the result, make the final decision and, when implementation is required, send a structured mission directly to Codex.
Codex would implement the change, run tests and return a structured execution report directly to the orchestrating Chat.
The user would still approve every important handoff.
Suggested UX
Inside a ChatGPT Project, I imagine actions such as:
- Send to Work
- Send to Codex
- Return result to Chat
- Request revision
- Approve and continue
The important part is that this should not simply forward the entire conversation.
Chat should be able to create a structured handoff containing only the relevant context:
- Mission objective
- Project context
- Scope
- Constraints
- Files or resources required
- Expected output
- Acceptance criteria
- Previous decisions relevant to the mission
The receiving agent would then return a structured result to the originating Chat.
For Codex, for example:
- Mission ID
- Status: PASS / FAIL / BLOCKED
- Files changed
- Tests executed
- Test results
- Build status
- What was implemented
- What remains
- Risks or blockers
- Commit or evidence when relevant
Chat could then automatically analyze that report and decide the next step with the user.
Why I Think This Would Be Valuable
The different tools already have complementary strengths:
- Chat: reasoning, decision-making, project orchestration and interaction with the user
- Work: long-running, multi-step analysis and production tasks
- Codex: software implementation, testing and development workflows
What seems to be missing is a native handoff layer between them.
Today, users often have to act as a “human clipboard” between agents.
This creates several problems:
- repeated copy/paste
- unnecessary context duplication
- risk of losing important constraints
- inconsistent versions of project information
- longer workflows
- more cognitive load for the user
A project-native handoff system could turn ChatGPT Projects into a much more coherent multi-agent workspace while still keeping the human in control.
Important: Keep It Optional
I would strongly recommend that this orchestration remain optional.
For many development tasks, my preferred workflow would still simply be:
Chat → Codex → Chat
Work should only be inserted when it brings real value, such as deeper research, document-heavy analysis, market studies, large deliverables or multi-step investigations.
The user should decide whether a mission goes directly to Codex or first goes through Work.
Example
For one of my software projects, my normal workflow could be:
Chat — Project Orchestrator
Defines the next product mission and acceptance criteria.
↓
Work — Optional
Performs deeper research or analysis when needed.
↓
Chat
Reviews the result and makes the product decision.
↓
Codex
Implements the approved solution and runs the tests.
↓
Chat
Receives the Codex execution report, validates PASS/FAIL and prepares the next mission.
This would preserve the strengths of each tool while dramatically reducing manual coordination.
In short, I think the opportunity is not only to connect Chat, Work and Codex, but to provide a simple human-approved orchestration and handoff system inside a ChatGPT Project.
I believe this could be especially useful for users managing long-running software, product or business projects with multiple specialized AI agents.