Multiplayer AI: When Agents Join the Team
Hey, it’s Samet 👋
For the last two years, every AI product has been obsessed with one idea: give each person their own personal assistant. One human, one agent, one workflow. It felt like magic, until you realized it doesn’t scale.
Your ChatGPT thread doesn’t talk to your teammate’s Claude thread. Your agent’s context dies the moment you close the tab. We built powerful single-player tools in a world that runs on teams.
That’s changing fast, and it’s the shift I want to unpack today: multiplayer AI, products designed not just for humans, and not just for agents, but for both, working side by side in the same shared space.
From Single-Player to Multiplayer
Dust, an agentic AI platform, just raised a $40M Series B specifically to build what they call “the multiplayer AI system for human-agent collaboration”. Their pitch is simple but sharp: most AI at work today is single-player, everyone gets their own agent, some tasks get faster, but the gains never compound across the team because there’s no shared context. Multiplayer AI flips that.
Humans and agents work in parallel, on the same workspace, with shared tools, shared artifacts, and shared goals, so intelligence compounds across the whole organization instead of staying trapped in one person’s tab.
Anthropic is seeing the same pattern. Their team describes the shift explicitly as moving “from a single-player to a multiplayer experience, where humans and agents work together as a team to achieve shared goals”. This isn’t a niche opinion anymore. It’s becoming the default framing for what agentic products should look like next.
Designing for Two Kinds of Users
Here’s where it gets interesting for us as PMs and designers: your product now has two user types living in the same interface, the human and the agent acting on their behalf.
Palantir’s product design team laid this out well at DevCon 6, focusing on how to present agent reasoning without causing information overload, and building clear attribution so users always know who made which decision, human or agent.
That attribution layer isn’t a nice-to-have anymore, it’s core UX. If your agent edits a doc or moves a task, your users need a visible log and an obvious undo path, the same way Figma shows you whose cursor is doing what.
Practical patterns worth stealing for your own roadmap:
Shared context layer: agents and teammates see the same knowledge base, not fragmented personal instances
Visible agent presence: show when an agent is “thinking,” executing, or idle, like a live cursor, but for AI
Dual-layer permissions: separate what an agent can access from who’s allowed to trigger it, especially for enterprise rollouts
Self-improving skills: let best practices discovered by one agent spread automatically across the team’s shared workspace
Frameworks like Dust, AutoGen, and CrewAI are already productizing pieces of this: agent teams that use tools, browse the web, and loop humans in when it matters.
Your Move This Week
If you’re building or managing an AI product right now, ask yourself one question: does my product still assume one human talks to one agent alone? If yes, you’re designing for last year’s paradigm. Start sketching what a shared, multiplayer surface would look like for your users and their agents, even a rough Figma mock of “who did what and when” will surface gaps in trust and attribution you haven’t thought about yet.
Multiplayer AI isn’t a feature. It’s the next operating model for how humans and machines actually get work done together, and the PMs who design for that shift early will have a real edge.
Talk soon,
Samet Özkale, AI for Product Power