All Episodes
Ep 1. Building multi-agent AI teams with OpenClaw
The first episode explores the fundamental vision of building an AI-Native organization by shifting from a single, generalized AI assistant to a collaborative team of highly specialized AI agents. It focuses on how to establish strict boundaries and roles for these agents using the Identity System, which is built on three pillars: the SOUL.md file (defining the agent's specific role, expertise, and communication style), the TOOLS.md file (defining their allowed capabilities), and the MEMORY.md file (tracking their ongoing experience). Ultimately, the episode highlights the Separation Principle, explaining how separating an agent's core personality from its tools ensures that introducing new capabilities doesn't rewrite who the agent is or cause them to step out of their specific department.
Ep 2. Deploying OpenClaw on a Mac Mini
The episode walks listeners through the initial prerequisites, such as setting up a new Google Gemini API, configuring budget limits, and creating a Telegram bot. It then dives into the technical installation process via the Mac terminal, specifically addressing how to overcome common hurdles like Xcode and Homebrew installation errors. Finally, the episode explains how to connect and pair the OpenClaw system with your Telegram bot, wrapping up with a critical security tip: configuring your Telegram channels so that your AI bots will only accept commands from your personal phone number, preventing unauthorized access.
Ep 3. Mission Control for Multiple AI Agents
It discusses how this dashboard eliminates the chaos of managing AI through basic chat interfaces by introducing a structured, Jira-like Kanban board for task management. The hosts explain how this setup allows humans to seamlessly assign and track tasks—without ever needing to learn complex AI commands or write prompts. Additionally, the episode covers how Mission Control helps establish a clear chain of command and hierarchy among the agents, ensuring that humans can easily manage and plan the strategy while the specialized AI agents execute the work
Ep 4. A Shared Brain for Sixteen AI Agents
Episode four explores the creation of a shared knowledge infrastructure to efficiently scale capabilities across a multi-agent workforce without duplicating effort. The hosts explain the "tool duplication problem," where manually updating each individual agent's files with new tools causes massive repetitive work and inevitable version drift. To overcome this, the episode introduces a master CAPABILITIES.md file that serves as a centralized operating manual detailing all installed tools and shared organizational rules. By using a "symlink" architecture rather than copying files into each workspace, any update made to this master document instantly propagates to the entire 16-agent workforce without synchronization delays. Finally, the episode highlights how an automated weekly "cron job" scans for newly installed tools and automatically updates the master capabilities file, ensuring you only have to teach the system about a new tool once for the entire organization to learn it
Ep 5. Building a Specialized AI Workforce with OpenClaw
Episode five focuses on the mechanics of the Automated Hiring Pipeline, detailing how the hiring-protocol.sh script seamlessly streamlines the onboarding of new AI agents by provisioning their database records, workspaces, and essential identity files. The hosts explore the system's strict approval flow, explaining that to maintain organizational integrity, the AI CEO (Winnie) must personally greenlight any new hires requested by others. A major theme of the episode is the critical importance of assigning Unique IDs (like UXD001), which act as a system-wide fingerprint that allows human managers to easily track an agent's tasks, hierarchy, and permissions anywhere in the system. Finally, the episode highlights crucial operational lessons learned from the UXD Talks blueprint, such as the necessity of deleting "ghost" test records to prevent gaps in the ID sequence, the requirement to manually sync the local filesystem with the database, and the importance of setting correct file permissions so agents can read each other's workspaces
Ep 6. Managing sixteen AI agents without going bankrupt
The final episode explores the critical operational and financial aspects of managing a multi-agent AI system, focusing heavily on model usage tracking and automated workflows. The hosts emphasize the importance of monitoring token consumption to prevent a "cost black box," specifically highlighting a "Leak Detection System" to catch expensive AI hallucination loops—such as an instance where the CEO agent, Winnie, burned through over 58,000 tokens in a single conversation. To optimize costs, the episode recommends matching the right AI model to the specific task, such as using the cheaper Gemini 2.5 Flash Lite for routine tasks while reserving Claude Opus for complex coding, which can yield about 75% in total cost savings. Finally, the hosts discuss the power of using "cron jobs" to automate daily recurring tasks like morning briefs and data syncs, while strongly warning about the financial dangers of misconfigured automation; for example, accidentally setting a daily cron job to run every minute can make the process 24 times more expensive and cause tasks to silently fail due to timeouts.