OpenClaw 2.0 touts 'multiplayer coding' with shared AI agents

OpenClaw 2.0 aims to shift teams from local harnesses to a shared coding agent that tracks work and coordinates tasks. Peter Steinberger says it can run across connected machines and cloud sessions, with pairing, WebVNC testing, and takeover support.

OpenClaw 2.0 touts 'multiplayer coding' with shared AI agents

TL;DR

  • OpenClaw 2.0: Shared coding agent used by a team to build the project over two months
  • Shared system: Shifts contributors from local harnesses to coordinated task tracking and ongoing work visibility
  • “Multiplayer coding” compute: Agents run on gateway, connected machines, or ~80 cloud providers; optional load balancing
  • Slack workflow: Feature discussion → OpenClaw returns a URL for pairing with the agent and testing results
  • WebVNC + portals: Agent gets a dedicated test computer; teammates can pair into the same session
  • Collisions and access control: Alerts on duplicate work, session takeover, default worktrees, private sessions, maintainer area with selected contributors

Peter Steinberger described OpenClaw as a shared coding agent that his team has used to build the project over the past two months. In a post accompanying an announcement that “OpenClaw 2.0 has arrived,” Steinberger stated that the project had moved contributors away from local coding harnesses and toward a shared system that tracks ongoing work and coordinates tasks.

Steinberger characterized the setup as “multiplayer coding” with compute available through connected machines and cloud sessions. According to his follow-up posts, agents can run on the gateway, on any connected machine, or through roughly 80 cloud providers. An automatic option can handle load balancing between those locations.

The workflow also extends beyond producing code changes for later local review. Steinberger described discussing a feature in Slack, then having OpenClaw return a URL for pairing with the agent and testing the result. The system includes WebVNC and portals that give an agent its own computer for testing, while allowing other team members to pair with the same session.

He also stated that the agent can alert contributors when multiple people are working on the same task and allow sessions to be taken over. Worktrees are enabled by default, although alternative choices can create conflicts. Steinberger additionally mentioned a mode for private sessions and a maintainer-focused area that can include selected contributors.

The project still appears to involve considerable configuration. Steinberger stated that an update reduced the number of configuration options from roughly 10,000 to fewer than 8,000, while acknowledging that testing every combination is impossible. He also noted that OpenClaw uses the same quotas as its command-line interface rather than creating separate ones.

Steinberger contrasted the shared model with local coding tools, which he described as feeling like “relics of the past.” That assessment reflects the project maintainer’s experience and does not establish that the approach is suitable for every development workflow.

Source: Peter Steinberger on X

Continue the conversation on Slack

Did this article spark your interest? Join our community of experts and enthusiasts to dive deeper, ask questions, and share your ideas.

Join our community