Onur Solmaz presents OpenClaw, an agent harness integrated with Kubernetes that leverages the Agent Client Protocol (ACP) and the ACPX CLI tool to automate and streamline agent interactions, particularly for managing large-scale open-source contributions. He highlights ongoing efforts to enhance agent interoperability, security, and dynamic provisioning across platforms, demonstrated through a Kubernetes-powered Slack error reporting system, and invites community collaboration to advance agent-driven development workflows.
In this talk, Onur Solmaz discusses his work on building an agent harness called OpenClaw and integrating open-source agents and frameworks on Kubernetes. He shares his background as a founding engineer at a startup and his journey with OpenClaw, starting from early interactions with ClaudeBot on Discord to becoming a maintainer focused on agent interoperability and orchestration. Onur emphasizes the importance of accelerating enterprise adoption of OpenClaw while addressing security concerns, noting that the project is still a work in progress.
Onur explains the development of ACP (Agent Client Protocol), a standard for agent-to-client interaction, which he chose due to its existing adapters for Codex and Cloud Code. He contrasts ACP with other protocols like MCP, which is for agent-to-agent communication. To facilitate agent interactions, he created ACPX, a CLI tool that acts as a Swiss Army knife for ACP, enabling agents to communicate and perform tasks programmatically. This tool supports workflows that automate repetitive tasks such as reviewing and refactoring code, helping to manage the influx of pull requests and feature requests in OpenClaw.
A significant challenge Onur highlights is managing the large volume of contributions and feedback from the OpenClaw community, which includes tens of thousands of stakeholders. To address this, he developed abstract workflows that automate the review and refactor loops for pull requests, aiming to reduce manual intervention and improve efficiency. These workflows are designed to handle superficial fixes automatically while escalating more complex issues to humans, thus creating a standard operating procedure for agent-driven development.
Onur also discusses the future vision of on-demand disposable agents that can be provisioned dynamically across platforms like Slack, Teams, and Discord. He notes the current limitations in multi-agent provisioning on these platforms and the need for standardized support to enable seamless interaction with multiple agents. To support this vision, he has been working on an open-source orchestrator called the Goal Operator, which manages agent deployment and interaction on Kubernetes, providing a scalable and flexible infrastructure for enterprise use cases.
Finally, Onur demonstrates the practical application of his work through a Slack-based error reporting system powered by Kubernetes-hosted agents. This system allows users to dispatch agents to debug issues in real-time, showcasing the power of giving agents full computational resources. He invites the community to explore and contribute to these open-source projects and offers assistance in deploying and integrating these tools for enterprise environments. The talk concludes with a call to embrace agent-driven workflows to enhance productivity and innovation in software development.
