Repository navigation
Feature Request: Let ChatGPT monitor and supervise active Codex tasks #33249
Description
Problem
ChatGPT can already hand tasks off to Codex, but it cannot see Codex's live progress or intermediate results.
In my workflow, ChatGPT can inspect a reference image, identify exactly what Codex modeled incorrectly, and generate precise correction instructions. But I still have to manually screenshot Codex's output, bring it back to ChatGPT, then copy the correction back into Codex.
This makes the user act as an API between two OpenAI products.
Proposed Solution
Ideally, ChatGPT should be able to:
- View the status and progress of a Codex task
- Inspect intermediate outputs
- Interrupt execution when something is clearly wrong
- Send correction instructions directly to Codex
Ideal Workflow
ChatGPT should be the supervisor. Codex should be the executor.
Motivation
The current ChatGPT → Codex handoff is missing half the feedback loop. ChatGPT has vision, reasoning, and context — it can evaluate Codex's work far better than the user copy-pasting between tabs. Without this supervision layer, users are forced to be the glue between two products from the same company.
This isn't about UI polish — it's about closing an architectural gap in the ChatGPT + Codex integration.
Activity
github-actions commentedon Jul 15, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
wangqj646-hub commentedon Jul 15, 2026
Note: This is not a duplicate of #32519. That issue focuses on a full bidirectional workflow (PRD versioning, task creation, change summaries). This issue is specifically about the real-time supervision and feedback loop — ChatGPT being able to see what Codex is doing mid-task and intervene when needed.
Also want to add an important point that's easy to overlook: not everyone is an engineer.
Most users can't write a precise technical requirement. They have a vague idea of what they want, and ChatGPT is the one that turns "I want something like this but nicer" into a structured task with clear steps. Without ChatGPT in the loop during execution, those users are left staring at Codex output they don't fully understand, with no way to tell if it's right or wrong — let alone how to fix it.
The real power of ChatGPT + Codex isn't just speed. It's that ChatGPT can interpret intent, Codex can execute it, and ChatGPT can verify the result. That three-step loop is what makes it usable for non-technical users. Without the supervision step, you've built a car with no dashboard — fast, but the driver can't see the road.