What is multiplayer AI?

Multiplayer AI is AI that several people use together: they share one session with the same AI agent, see each other's requests and its answers as they happen, and can all steer it, instead of each person working with AI in a private chat. The term spread in 2026. Y Combinator put multiplayer AI in its Fall 2026 Request for Startups, Anthropic teaches it in a Claude Academy course, and vendors from Dust to ThoughtSpot use it. They don't all mean the same thing. Three meanings are in use, and this page sorts them out.

Key takeaways

• Multiplayer AI usually means several people and one AI in one shared session, the way Google Docs made documents multiplayer.
• Two other meanings are in use: many AI agents working together, and a team sharing AI context or memory while each person still chats alone.
• The two biggest consumer group chats closed in 2026: ChatGPT group chats stopped taking new chats on July 9, and Microsoft's consumer Copilot Groups retired from August 18.
• What remains is mostly at work: an AI in a team channel, such as Claude Tag or Microsoft 365 Copilot in Teams, or a shared workspace built for it, such as Poly for building software.

Why is it called multiplayer?

Most AI is single-player. A chat belongs to one person, and to share what the AI did, that person copies the answer into Slack or a doc. The rest of the team gets it second-hand and can't steer it. Software went through this before: Google Docs replaced emailing Word files, and Figma replaced passing Photoshop files around. YC's request makes the same comparison, asking for agents that teams share in live sessions, where they can watch it work, redirect it and hand it off. Sergey Karayev's Multiplayer AI Manifesto (September 8, 2026) puts the goal in one line: "Teammates can participate in the same agent session." Y Combinator made the same case in its Fall 2026 requests for startups: YC's Multiplayer AI request, explained.

What are the three meanings?

1. Several people, one AI session

This is the strict meaning, and the one this page uses. Superconductor's explainer (September 9, 2026) offers a simple test: can a teammate talk to the agent you are already working with? Claude Academy contrasts single-player AI, a "super suit" for one person, with a shared room. Salesforce Ben's guide to multiplayer agents (September 30, 2026) describes agents that several users and agents can contribute to at the same time.

2. Many agents working together

ThoughtSpot's explainer (September 22, 2026) defines multiplayer AI as organizing multiple AI agents around a shared goal, with people overseeing. Most people call that multi-agent AI. Adding agents doesn't put more people in the conversation; multiplayer is not multi-agent, though a product can be both.

3. Shared team context

Elvex (June 2026) means a shared team environment for agents, context, workflows and data. OrgLogic calls it AI a team uses together rather than one person at a time. PromptQL (September 4, 2026) means a shared agent with memory everyone can reach, and Dust, which raised a $40 million Series B in May 2026, means agents and people sharing context across a company. Shared context matters, because it stops every person's AI from relearning what a teammate already taught it. But people may still each work in their own chat.

What makes AI truly multiplayer?

An AI that several people can mention in a group chat is the simplest form. A fuller version has six properties:

1. One session everyone sees live. Requests, answers and the AI's work stream to everyone as they happen.
2. Anyone can steer. Any member can ask, answer the AI's questions, redirect it or stop it.
3. Every step has a name. Each request, approval and result says who did it.
4. A say before the AI acts. Anyone in the session can approve or reject what the AI wants to do.
5. A place to talk the AI doesn't read. People argue about the plan without the argument becoming the prompt.
6. One shared result. The work lands in one place everyone can see, not in each person's chat history.

Superconductor's post lists its own principles and admits that no platform, Superconductor included, meets all of them yet.

Where can you use multiplayer AI today?

• At work, in team chat: Claude Tag puts one Claude in a Slack channel for Claude Team and Enterprise organizations. Microsoft 365 Copilot answers in Teams group chats, though only people with a Copilot license can ask it questions. Dust lets several people work with the same agents in one conversation.
• For building software: Slack Code opens a code channel around a coding agent. Poly, AQ, Superconductor, HumanLayer and Zed's Delta let several people steer one live coding agent.
• With friends: Poe's group chats put several people and several AI bots in one thread, and Jenova has a multiplayer group chat.
• Closed: ChatGPT group chats, piloted from November 2025, have accepted no new chats since July 9, 2026 (OpenAI). Microsoft retired consumer Copilot Groups from August 18, 2026. What happened to ChatGPT group chats.

Each tool is compared, with prices, in Multiplayer AI tools.

Where does Poly fit?

Poly (usepoly.co) is multiplayer AI for building things, in the strict sense: a room where a team works with one AI together. Everyone sees the same live timeline, any member can approve a change before it happens, every turn has a receipt naming who ran it and what it cost, and the group chat is one the AI never reads. Friends use it too, to plan a trip in one shared chat. What is Poly?