We built OpenClaw Ultron to replace 20 people at our company | E2246

What It Covers

Jason Calacanis and team demonstrate OpenClaw Ultron, their AI agent built to automate tasks across their venture firm and podcast production company. Oliver Korzen showcases the dashboard, cron jobs, and skills developed over two weeks. Guest Alex Cheema from Exo Labs discusses running frontier AI models locally on consumer hardware like Mac Studios to maintain data sovereignty and avoid vendor lock-in.

Key Questions Answered

• •Production Automation Timeline: After two weeks of building OpenClaw Ultron, Oliver estimates 60% of his thirty weekly production hours will be automated within thirty days. The system handles guest research, outreach, calendar management, and sponsor identification through scheduled cron jobs. This demonstrates rapid implementation velocity where one skill gets built approximately every 1.5 days, suggesting 200 total company skills could be deployed within months at current pace.
• •Dashboard-First Development: Building a visual dashboard should be step one when deploying OpenClaw, not an afterthought. The dashboard connects to OpenClaw's backend to display memory files, preferences, cron jobs, skills, and schedules visually rather than querying through chat interface. Oliver created his dashboard by screenshotting Alex Finn's YouTube video and having OpenClaw replicate it, demonstrating how vibe coding accelerates custom tool development for knowledge workers.
• •Local AI Hardware Economics: Two Mac Studios with 512GB memory cost approximately $20,000 and represent the cheapest way to run frontier models like Qwen 2.5 locally. Apple's RDMA support via Thunderbolt 5 cables ($50) enables low-latency memory sharing between devices, creating one unified GPU. This approach eliminates per-token costs, prevents vendor lock-in, and ensures models don't change unexpectedly, with enterprise customers now clustering over 100 Mac Minis for various workloads.
• •Self-Optimization Capability: OpenClaw runs a self-optimization cron job Monday through Friday from 3-5AM, analyzing all files, cron jobs, and skills to identify improvements. At 8AM it delivers five actionable recommendations without executing changes. Examples include detecting timezone bugs in calendar systems and identifying cron scheduler issues causing skipped jobs. This creates a continuous improvement loop where the AI audits and enhances its own infrastructure daily.
• •Automated Competitive Intelligence: A cron job monitors approximately twenty competitor podcasts using YouTube API and Podscribe, extracting sponsor information from timestamps. It cross-references findings against the Pipedrive CRM to identify which sales rep owns each relationship, then sends daily Slack messages flagging new sponsors or unowned opportunities. This replaces manual research that previously required dedicated staff hours, running continuously 365 days annually with perfect consistency.
• •Human-in-Loop Workflow Design: Guest booking remains intentionally human-in-loop despite automation capabilities. The system sources five guest ideas daily at 7:45AM, performs deep research using separate skills, and presents recommendations rather than executing end-to-end. This reflects strategic trust boundaries where subjective decisions about guest quality, personality fit, and show chemistry still require human judgment, even as objective tasks like scheduling and research get fully automated.