Feedback for Improvement:
I have a suggestion for Codex/ChatGPT to improve complex task automation, especially in Chrome and web-based workflows. I often want to give one high-level instruction that requires many steps across different webpages, Google Docs, Gmail, Sheets, Drive, and other sites. Right now, even when Codex understands the task, it can feel slow because it has to work through each step sequentially.
A helpful improvement would be for Codex to act as a supervising coordinator that can delegate subtasks to other integrated agents, including Google-integrated Gemini or similar browser/workspace agents when available. For example, Codex could plan the overall workflow, then ask a Google-integrated agent to handle Google Docs, Gmail, Drive, Calendar, or Sheets-specific tasks while Codex continues handling research, coding, file work, verification, or other browser steps in parallel.
This could work especially well if Codex could communicate structured instructions to these agents instead of trying to manually control their sidebar UI. Codex could say something like: “Gemini, summarize this Google Doc and extract action items,” or “Use Google Workspace context to find related emails and calendar events,” then receive the result back and verify it before continuing.
It would also be useful if Codex could delegate simpler or repetitive work to lower-cost worker agents while keeping a stronger model in charge of planning, quality control, user approvals, and safety. Clear permission boundaries, visible progress updates, and final verification by Codex would make this feel trustworthy.
The current experience is powerful, but complex workflows can take a long time when handled one step at a time. I would love to see Codex support supervised parallel task execution with integrated agents like Google’s Gemini or other specialized assistants, so users can describe the goal once and let the system coordinate the boring multi-step work efficiently.