What is multiplayer AI? How teams of people and AI agents share context and work together

What is multiplayer AI?

Multiplayer AI is a way of working where a group of people and a group of AI agents work toward the same goals from the same context. The agents are not private assistants that each person talks to alone. They are AI teammates: each has a role, can see what the team has decided, hands work to people and to other agents, and stops to ask when something needs a human.

The name comes from games. In a single-player game the world exists for one person and resets when they leave. In a multiplayer game everyone is in the same world, sees the same state and plays a part. AI works the same way:

It's worth saying what multiplayer AI is not. It isn't a shared chat where five people type to one bot, and it isn't a pile of agents running unwatched in the background. It's closer to a team.

Why does multiplayer AI matter now?

Two things changed. First, AI moved from answering questions to doing work. Agents can now research, draft, check and follow up over many steps, often connected to the tools a team already uses through standards such as the Model Context Protocol. Once an agent does work instead of answering a question, that work has to fit into how the team operates.

Second, most teams have already tried the single-player version and felt its limits:

• Everyone re-explains everything. Each person keeps their own chats with their own context, and none of it is shared.
• Good answers get stranded. A useful analysis lives in one person's chat history and never reaches the person who needed it.
• Someone becomes the router. One person, often the founder or a team lead, carries every AI result to the next step by hand.
• Nobody knows what the AI did. When something goes wrong, there is no shared record of what was asked, what came back and who approved it.

What makes multiplayer AI work?

From what we've learned building toward this, five things matter more than which model you pick:

• Shared context and memory. Goals, customers, decisions and what the team has learned, kept in one place every person and agent works from, and updated as work finishes. This is the foundation. Without it you have several single-player sessions running side by side.
• Clear roles and ownership. Each AI teammate has a job, such as research, analysis or drafting, and each outcome has a person who owns it. "The agent did it" is not an answer a customer accepts.
• Handoffs that carry context. When work moves from an agent to a person, or from one group to the next, the reasoning and evidence move with it, not just a one-line summary.
• Humans in the loop by design. Decide up front what runs on its own, what waits for a yes and what only a person decides. Publishing, spending money and changing anything customers see should wait for approval at first.
• Permissions and privacy. Agents should see only what their role needs, and you should know where your team's context is stored and who can read it. Shared doesn't have to mean public, or uploaded somewhere you don't control.

If you want the structure underneath this in more detail, I wrote about it in what a company harness is.

What do teams do with AI teammates?

A few examples of an AI agent team working next to people. None of them needs a big company.

• Customer feedback. One agent gathers support messages and reviews each week, another groups them into themes, and a product person decides what to act on. The themes and the decision go back into the shared context, so next week starts from there.
• Research before a decision. Before a pricing change, one agent looks at competitors, another pulls together what customers have said about price, and the founder makes the call with the evidence in front of them.
• Launching something. A person writes the core message, an agent drafts versions for each channel, another checks them against what was decided, and nothing is published until someone approves it.
• Onboarding a new hire. Instead of a week of "ask Sam", the new person and an AI coworker work from the same context as the rest of the team: past decisions, open questions and who owns what.

The pattern is the same each time: agents gather, draft and check, people own the outcome, and the result feeds back into what everyone works from.

What are the pitfalls of multiplayer AI?

• Context sprawl. If every person and agent writes notes, summaries and memories in different places, the shared context turns into a pile nobody trusts. Keep one source of context, and prune it the way you would a shared document.
• Agents stepping on each other. Two agents rewrite the same draft, redo the same research or act on each other's half-finished work. Clear roles and one owner per piece of work fix most of this.
• Errors that compound. One agent passes a wrong assumption to the next, and three steps later it looks like a fact. Handoffs with the evidence attached make it visible, and approvals catch it before it reaches customers.
• Data leaving your control. Shared context is valuable, which is exactly why you should care where it goes. Check which services store it, who can read it and whether it is used for anything beyond your work.
• Too much autonomy too early. It is easier to move work from "needs approval" to "runs on its own" once it has earned trust than to pull it back after a mistake.

How do you get started with an AI agent team?

Start small, with the tools you already have:

1. Write down the shared context. Goals, customers, recent decisions and how you like things done, in one place both people and agents can read.
2. Pick one outcome. Not "use AI more", but something concrete, like a weekly summary of customer feedback. Give it a human owner.
3. Give each agent one role. One gathers, one drafts, one checks. A few clear roles beat one agent that does everything.
4. Decide what needs a person. Write down what may run alone, what waits for approval and what is human only, before anything runs.
5. Make the handoff explicit. Say what gets passed on, to whom, and with which evidence.
6. Review every week. Look at what worked and what got stuck, and add what you learned to the shared context.

You'll hit the limits of doing this by hand fairly quickly. Context drifts across chats and documents, handoffs depend on someone remembering, and approvals live in your head. That's the part a multiplayer AI tool should hold for you. If you run a small business and are deciding what to hand off first, AI agents for small business goes into more detail.

Where does Koru fit?

Koru Squads is the multiplayer AI workspace we're building now. It brings humans and AI agents together around shared goals, shared context and coordinated work. It is in development and not available yet.

The idea is to start with a goal, then build the squads to run it. A squad is a persistent group made up of whatever the outcome needs: people, agents, or both. Squads work from shared company context, own outcomes, hand work to each other with the context retained, and come to you at clear human checkpoints. We're designing controls that separate what runs automatically, what needs approval and what stays with people. You are not managing a collection of AI bots. You are building a team.

What you can use today is Koru Workspace, our free app for Windows and macOS, with mobile and Linux coming soon. It keeps your work on your device in a local database, and your devices can sync directly with each other, end-to-end encrypted. It doesn't do multiplayer AI. That's what Squads is for.

If that's how you want to work with AI, join the Koru Squads early access waitlist. We're bringing early users in gradually, and joining early is a way to help shape how it works. If you want a private place for your own work in the meantime, download Koru Workspace.