The Multiplayer AI Manifesto

The manifesto names a real shift in agent work: individual AI chats create context tax, security gaps, and lost organizational learning; shared cloud sessions turn agents into collaborative work surfaces.

Logged at IST: 2026-09-05 11:30 IST

What it is: Sergey Karayev sharing a manifesto for “multiplayer AI”: cloud-hosted, team-joinable agent sessions that live next to the work instead of inside private single-user chat boxes.

Gist: The manifesto argues that work moved from collaborative software into siloed AI chats. People ask agents private questions, copy results into Slack or email, and force teammates to re-prompt another model. That creates a context tax, prevents teammates from learning from each other, and leaves organizations with laptop-bound agents that are hard to secure, audit, or resume.

The five principles are crisp: never copy and paste, work with the door open, continuously improve, people are not routers, and nothing starts from scratch. In practice, that means shared sessions attached to artifacts, agents reachable from multiple work surfaces, reusable skills generated from corrections, benchmarks created from repeated workflows, and resumable context for every project, PR, doc, or plan.

The security framing is the useful counterweight to the product pitch. Multiplayer agents need cloud sandboxes, allow-listed service access, permission intersections when new people join a session, provider agnosticism, and governance records that can answer who requested the work, which agent ran, what data it touched, and what changed.

Newsletter angle: Strong companion to the agent-operating-surface lane: the next step after “everyone has a copilot” is shared, audited, resumable agent work. The hard part is not chat UX. It is permissions, durable context, learning loops, and avoiding humans as message routers.

Embedded source

{{}}