AI Agents Just Became Multiplayer. Here's How to Build Them.

Kylon: Architecting Multiplayer AI Workspaces

Creator Nate Herk introduces Kylon, a collaborative workspace designed to address the fragmentation of individual AI operating systems (AIOS). Initially hesitant about the tool's value, Herk demonstrates how Kylon facilitates "multiplayer AI," where teams of humans and specialized agents collaborate within a unified environment.

Herk contrasts the current "single-player" landscape—siloed AIOS with unique skills and integrations—with Kylon's shared vision. The core challenge is duplicated effort and data syncing issues across tools like Google Drive and ClickUp (e.g., forgetting to mark tasks complete). The proposed solution is a central hub where agents are introduced to the team, communicate directly, and access shared tools. The platform's interface mimics a hybrid of Slack and ClickUp, organized into channels and rooms for specific projects (e.g., "AIS Media," "AIS Instagram"). Users can interact with agents and teammates in group channels, develop custom applications (dashboards, trackers, approval pipelines) and automations (scheduled tasks) directly in the workspace, and structure work streams into threads for focused, asynchronous collaboration.

Kylon provides significant architectural flexibility in how agents are powered. Users can deploy Kylon-hosted agents (which intelligently route tasks to models like Opus 5.5 on the "Max" tier) or connect their own Claude Code and Codex subscriptions. A key distinction is that local subscriptions run on the user's machine and require it to be on, while Kylon cloud agents run 24/7. Agents have intelligence tiers (Fast, Pro, Max) to optimize cost and performance. The platform charges per workspace, not per user, allowing teams to invite members without incurring additional base costs. Credits are consumed by Kylon-hosted agents, while connected subscriptions run on the user's own dime. The system provides granular visibility into spending per person and per room.

Herk advocates for a phased adoption strategy, starting with a single business unit to find quick wins. Users import their existing AIOS context (skills, memory) into Kylon and connect essential tools (Gmail, GitHub, Notion) via OAuth or API keys. A salient example is an automated Instagram channel where agents research, ideate, create carousels, and manage an approval pipeline, posting directly to the account. Another example is a weekly commitment tracker that syncs with ClickUp to improve team accountability by pinging members about overdue tasks. Herk emphasizes that Kylon is not intended to replace existing tech stacks like ClickUp, but rather to connect to them and enhance workflows with persistent, collaborative agents.

Key Takeaway

Kylon represents a significant step toward operationalizing AI within teams. It abstracts the complex technical overhead of building custom multi-agent systems, offering a ready-made interface for human-agent collaboration.

The central insight is that the value of an AIOS grows exponentially when it is shared.

By allowing agents to interact and access a collective source of truth, Kylon enables teams to move beyond individual efficiency gains toward a genuinely collaborative, AI-native operating system, effectively creating a "Grokbot for your entire team."