My OpenClaw & Qwen 3.5 AI Agent Filled My NCAA Bracket While I Was in Meetings — Live Demo

A live demonstration of 'Locus,' an AI Chief of Staff agent built with OpenClaw and Qwen 3.5 35B running locally on Mac Studio M3 Ultra, autonomously researching and filling out an NCAA bracket while the creator was in meetings. The agent showcases advanced agentic capabilities including web research, data analysis, and decision-making without human intervention. This demonstrates the practical application of local LLMs and agentic frameworks for real-world productivity tasks.

Key Points

• •OpenClaw framework enables autonomous agent behavior with tool use and decision-making capabilities
• •Qwen 3.5 35B model runs efficiently on local hardware (Mac Studio M3 Ultra) without cloud dependencies
• •AI agents can perform complex multi-step tasks like bracket research and selection autonomously
• •Local LLM deployment provides privacy and control compared to cloud-based AI services
• •Agent autonomy allows task completion during unavailability (meetings) without human oversight
• •Web research integration enables agents to gather real-time data for informed decision-making
• •Mac Studio M3 Ultra provides sufficient computational resources for running large language models locally
• •Agentic AI systems can handle domain-specific tasks (sports analytics) with minimal configuration

Workflow Diagram

Concepts