The video explains that to create jobs for OpenClaw AI agents, you should treat them as employees handling recurring roles rather than just assistants for one-off tasks, enabling scalable delegation of ongoing business processes. Brian Castle demonstrates his custom system for managing agents, where tasks are linked to modular “skills” (markdown-based instructions), and agent outputs are organized and reviewed using integrated tools, streamlining workflow improvement and business documentation.
The video addresses the most common question about OpenClaw: what kinds of jobs can you create for AI agents, and how should you structure their work? The creator, Brian Castle, emphasizes that while specific tasks will vary by business, the key is to shift from thinking of agents as personal assistants for one-off tasks to treating them as employees filling recurring roles. This mindset allows for true scalability, as it removes the business owner as the bottleneck and enables the delegation of ongoing responsibilities rather than just ad hoc assignments.
Brian explains that the foundation for creating jobs for OpenClaw agents is identifying recurring needs within your business. Unlike hiring human employees, where you need enough work to justify a salary, AI agents can be assigned even a few recurring tasks without significant cost. This lowers the threshold for delegation and allows business owners to start offloading work much sooner, freeing up their time for higher-value activities. He suggests looking for both tasks you already do and want to delegate, as well as missed opportunities that could be addressed with agent support.
To operationalize this, Brian has developed a custom system called BMHQ, a Rails app that acts as a mission control dashboard for managing multiple agents. This system includes a Kanban-style task board, scheduling features, and the ability to assign recurring tasks to specific agents. Each task typically points the agent to a particular “skill”—a set of instructions or scripts stored as markdown files—which defines the process for completing the job. This approach allows for easy refinement and improvement of processes over time, as updating a skill updates the workflow for all future tasks.
Skills are central to Brian’s system. They are organized as folders containing markdown files and supporting scripts, serving as operating manuals for each job. By storing and continuously improving these skills, Brian ensures that his agents follow consistent, up-to-date processes. He collaborates with Claude Code to refine these skills, making it easy to iterate on workflows and adapt to changing business needs. This modular approach also means that instructions are not hardcoded into individual tasks, making the system more flexible and maintainable.
Finally, Brian discusses how agent outputs are managed and reviewed. He built a custom markdown editor and viewer called Brainown, which integrates with Dropbox to store and access all artifacts produced by agents. Agents send links to these markdown files via Telegram, making it easy for Brian to review their work and provide feedback. This system of organizing business knowledge and outputs in markdown files streamlines collaboration and documentation. Brian concludes by reiterating that the real value of OpenClaw lies in treating agents as employees, focusing on building systems that enable delegation and scalability, and using tools like Claude Code to continuously improve business operations.