I know OpenAI already has many of the building blocks for this:
- Agents SDK with agents, handoffs and orchestration
- Workspace Agents with persistent context, tools and scheduled work
- Codex for autonomous software development
- Tasks / automations
- Computer use
- MCP, apps and external integrations
So my question is not whether it is technically possible to build a multi-agent system with OpenAI today. Clearly, it is.
What I haven’t found yet is a first-class OpenAI product/interface for creating and managing a persistent team of individual AI agents, similar to the product experience recently introduced by Grok Bot.
For example, imagine opening ChatGPT and having a permanent team such as:
- Manager / Chief of Staff
- Software Engineer
- Product Manager
- QA Engineer
- Marketing / Growth
- Researcher
Each agent would have its own identity, role, instructions, memory/context, tools, workspace and recurring responsibilities.
The important distinction is that these wouldn’t simply be temporary sub-agents spawned for a single task. They would be persistent AI teammates that remain available over time, accumulate context about their responsibilities and can collaborate with or delegate work to other agents.
Ideally, there would also be a dedicated interface where I could:
- see all of my agents and their current status
- open a conversation with any individual agent
- create group conversations involving multiple agents
- assign responsibilities and recurring routines
- see what each agent is currently working on
- allow a manager agent to delegate tasks to specialist agents
- review outputs, approvals and completed work
- maintain the same team across days, weeks and projects
In other words, OpenAI seems to already have most of the primitives required for this.
What appears to be missing is the product layer that turns those primitives into something that feels like managing an actual AI team rather than manually orchestrating agents, prompts, workflows and API calls.
Grok Bot is a useful reference here because it packages this concept into a very understandable interface: create individual bots, give each one a role, let them work persistently, create routines and allow multiple bots to collaborate.
Is OpenAI working toward something similar as a native ChatGPT experience?
Perhaps an evolution of Workspace Agents into something like a Personal AI Team / Agent Workspace, eventually available beyond Business/Enterprise?
And if not, is this a direction other developers here would find useful?
Personally, I think the next major step for AI assistants may not simply be having a more capable single assistant.
It may be moving from:
“I have an AI assistant.”
to:
“I have a persistent team of AI specialists that works with me.”