Tech • AI • Robotics • Game
ClawLabs: Building Open Source Together | DevDay 2026
7/10TL;DR
OpenClaw has expanded from a personal agent workflow into a collaborative, open-source platform for persistent software agents, and its new OpenClaw Enterprise control plane adds multi-tenant governance, security boundaries, and deployment controls for organizations.
KEY POINTS
From solo terminal work to shared agent systems
Agent-driven development began in command-line workflows, where multiple coding sessions were managed manually on local machines. As models improved, less direct intervention was needed, pushing work toward Codex Desktop, concurrent sessions, and then distributed compute across several machines and cloud test boxes. That scale solved throughput limits but exposed a collaboration problem: teammates still depended on one person to mediate access to the agents.
OpenClaw’s role as a team workspace
OpenClaw is positioned as an open-source platform for persistent agents that works across tools, devices, and teams. It supports shared visibility into sessions, prompts, and outputs, allowing collaborators to inspect, steer, and reuse one another’s work directly rather than relying on pull requests alone. The project is owned by the independent nonprofit OpenClaw Foundation, with support from OpenAI.
A multiplayer model for software development
The platform is designed around a team server where participants can see each other’s sessions, prompt active agents, and collaborate in real time. That approach is meant for groups with the same access rights, especially open-source teams and internal engineering groups. The shift from isolated agent use to shared sessions was described as increasing development velocity and reducing interpersonal friction because prompt history makes intent visible earlier.
Prompt history becomes part of code review
A central problem in AI-assisted coding is that a short prompt can generate thousands of lines of code while the reasoning behind those changes remains hidden on a local machine. In practice, teams often understand what changed in a pull request but not why. OpenClaw’s shared-session model addresses that by exposing the conversation around a change, helping teammates catch mistakes, understand trade-offs, and avoid conflicts before they escalate.
Dashboards, reports, and “jellyware”
The platform treats many internal tools as continuously evolving artifacts produced by agents. Dashboards track items such as CI duration, pull requests, issues, merge activity, and code quality drift through a “slop meter” that monitors overgenerated code. These tools can be updated conversationally, blurring the line between software and rapidly iterated agent-built outputs described as “jellyware.”
Persistent loops and long-running goals
OpenClaw supports both recurring automation loops and longer-term goals. Monitoring loops can scan Discord, GitHub, and X for bug reports, reproduce issues in fresh test boxes, generate fixes, run validation, and send the result for a second model-based review before changes land. Separately, long-running goals can chip away at architectural migrations over weeks; one example cited had run for 18 days and landed more than 100 pull requests.
Meeting agents and shared operational memory
The platform also extends beyond code into team operations. Meeting software was replaced with an agent that joins Discord, summarizes discussions, and can answer follow-up questions using a persistent shared identity and memory. Reports can track developer activity, while a daily automatically generated internal newspaper summarizes merged work, combining practical coordination with lightweight team culture.
Launch of OpenClaw Enterprise
OpenClaw Enterprise, or OCE, introduces an open-source control plane that makes OpenClaw multi-tenant and adds governance for organizational deployment. It is presented as a vendor-neutral layer for running and managing agents at scale, similar in ambition to how Kubernetes standardized container deployment. The project started internally at OpenAI and was donated early to the OpenClaw Foundation, with continued development alongside partners including Red Hat and NVIDIA.
Security controls for agent deployment
OCE’s design centers on strong isolation and auditability. Deterministic code runs in separate containers and clusters from agent execution so agents cannot modify their own runtime. Sandboxing limits network access, writable files, and repository permissions, while model-based auto-review can block dangerous actions before execution. Agents also run under dedicated identities and service accounts rather than impersonating users, allowing them to be independently controlled, audited, and disabled.
Scale inside OpenAI
Internal deployments were said to have grown to more than 400 Claws operating continuously. These agents handle coding, experimentation, data analysis, issue triage, and security escalation. A demonstration showed policy-based presets for launching an agent with the Codex harness, read-only repository access, Linear integration with approval controls, and Slack connectivity, alongside another persistent software agent modifying the enterprise dashboard itself.
CONCLUSION
OpenClaw reflects a broader shift from single-user coding assistants to shared, persistent agent systems embedded in team workflows. With OCE, the main challenge moves from raw model capability to governance, security, and coordination at organizational scale.
Ask a question