GPT-6 Pro plan in Chat, handed to Codex

I want to solve [specific problem]. Use the connected [GitHub repository and branch / logs / database] to read the relevant code, recent PRs, and records. Start your response with a brief list of the files and sources you actually read. If you can’t access or verify the requested branch, say so. Analyze the problem against [my goals and constraints], propose a plan, and tie your conclusions to specific files or records. Flag missing information rather than filling the gaps with guesses.

List decisions that need my input separately. Once we agree on the plan, turn it into a task brief for Codex with the scope of changes, supporting evidence, and acceptance criteria. Don’t modify code or data yet.

GPT-6 Pro in regular Chat doesn’t use the shared Work and Codex allowance. But it has its own limits: currently, $100 Pro gets 50 messages per week shared with GPT-5.6 Sol Pro, while $200 Pro gets 200 GPT-6 Pro messages per week. OpenAI’s notes are at https://help.openai.com/en/articles/20001354-gpt-5-6.

If I just ask “How can I improve my project?”, it doesn’t know what I recently changed, where users get stuck, or what errors the server is logging. If I want Pro to help with planning and solution design, I need to give it the information those decisions depend on. For a slow feature, that means the relevant code, recent PRs and actual logs.

Connect those sources directly to ChatGPT: use the existing GitHub connector for your repository, and a read-only MCP for your own logs or database where needed. That’s the division of work I want to try: let Pro work through the plan in Chat using real project data, agree on what needs doing, then hand it to Codex to implement and verify.

Once your sources are connected, try it with a real problem.

Plus users can also discuss a plan with a model available in regular Chat before handing it to Codex. Chat has a separate allowance; switching to Work puts you back on the allowance shared with Codex.