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

Summary

Jason Calacanis and Lon Harris host Oliver Cororsan to demo OpenClaw Ultron, an AI agent project automating internal podcast and venture tasks at TWiST/LAUNCH. Alex Cheema of ExoLabs discusses running frontier LLMs locally on consumer hardware, especially clustered Apple Mac Studios and Mac Minis, and argues open-source models like Kimi K2.5 are closing the gap with closed models. Ryan Yanneli of NextVisit AI presents the company's AI medical scribe for behavioral health after winning the Gamma pitch competition. The conversation also covers AI security risks, productivity leverage, and entertainment/media commentary.

• OpenClaw Ultron aims to automate 100-200 skills across TWiST/LAUNCH.
• Oliver Cororsan demonstrates dashboards, memory files, cron jobs, and skills for the agent.
• Alex Cheema explains ExoLabs' local AI hardware clustering and enterprise model.
• Apple silicon, especially Mac Studios and Mac Minis, is presented as cost-effective for local LLM inference.
• Open-source models like Kimi K2.5 are seen as closing the gap with closed models.
• NextVisit AI wins the Gamma pitch competition with an AI scribe for behavioral health.
• Prompt injection and skill security remain unsolved risks for AI agents.
• The episode ends with a discussion of HBO's Industry and fintech/short-seller themes.