Feature request: “My AI Team” — persistent named specialists with roles and avatars
What variant of Codex are you using?
ChatGPT Work in a web browser (Codex Work mode)
What feature would you like to see?
I would like to propose a feature called “My AI Team” for ChatGPT Work / Codex Work.
Many users work with several long-running chats at the same time. Each chat gradually develops its own responsibilities and context, but visually the chats look almost identical. This makes it difficult to remember which chat is responsible for which task, and users may accidentally write to the wrong chat or repeatedly explain the same requirements.
The proposed feature would allow a user to create persistent digital team members and:
- choose a name and avatar for each team member;
- define their role and responsibilities;
- specify what they should and should not do;
- set the expected working style and quality standards;
- preserve each member’s own instructions, working history, and context;
- display the member’s name, avatar, specialization, and current assignment in the sidebar;
- edit the instructions, pause the member, or archive the role.
For example:
- Nyusya — questionnaires and client materials;
- Lyusya — website content;
- Marusya — images and visual design;
- Petya — testing and quality control.
This is more than adding avatars to chats. Names, faces, roles, and persistent context would allow users to build and manage a recognizable team of specialized AI assistants.
The feature could reduce repeated explanations, prevent tasks from being mixed between chats, and make parallel work much easier to manage. It could be especially valuable for business owners, managers, older users, non-technical users, and anyone working on several long-term projects.
If this request belongs to another OpenAI product area, please route it to the team responsible for the ChatGPT Work / Codex Work interface.
Additional information
I previously managed a team of approximately 50 people, so I naturally began organizing my ChatGPT chats as employees with different responsibilities.
Today I accidentally wrote a sentence in the wrong chat. Because the assistant responsible for that task had already been given the name “Nyusya,” I immediately noticed the mistake and remembered where the message belonged.
This showed me that names and avatars are not merely decorative. They act as visual and mental landmarks: the user immediately recognizes who they are working with, what has already been discussed, and what that particular specialist is responsible for.
2 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I would like to add another essential part to the “My AI Team” concept: controlled learning and workspace cleanup for persistent AI employees.
At present, working on a long-term task can become very difficult. The user may carefully explain a requirement, correct the assistant several times, and finally approve the correct result. However, in a later generation, the assistant may lose that instruction and repeat the same mistake.
For example, while creating a printable questionnaire, I repeatedly explained that each checkbox must appear directly before the capital letter at the beginning of the first line. After several unsuccessful attempts, the assistant finally understood and produced the correct result. This learned rule should become part of that employee’s confirmed professional knowledge and should not disappear during the next task.
There is also a serious workspace-management problem. Long chats accumulate many incorrect images, drafts, failed generations, outdated files, and abandoned attempts. These materials remain visible, make it difficult to find the approved version, clutter the workspace, and may continue to influence later work. The user often has no simple way to remove an individual failed result or clearly tell the system: “This version is wrong. Do not use or learn from it.”
A persistent AI employee should therefore have:
The most important principle is that the AI employee should learn primarily from user-approved results, not from every intermediate attempt.
Professional status could then reflect genuine experience:
Progress should be based on confirmed successful work, learned rules, and fewer repeated mistakes—not simply on the number of messages.
This would make long-term collaboration much more practical. Users would be able to train an AI employee, preserve valuable experience, remove incorrect material, and maintain a clean workspace instead of repeatedly starting over and accumulating unusable results.