AI West talk - Agents are just while loops

A talk by Luke from Helix ML that cuts through the hype around AI agents with a straightforward technical explanation. The core idea? AI agents are simply language models that can use tools, wrapped in a while loop that continues until the task is complete. The talk includes: A brief explanation of while loops and how they form the foundation of agentic systems Demo 1: Building a functional Tic-Tac-Toe game using an AI coding agent Demo 2: Using an AI research agent to analyze academic papers comparing multi-agent vs single-agent performance Demo 3: Running similar capabilities locally with open source models for privacy and control Interesting technical points include how vision models paradoxically handle spreadsheet data better when it's converted to images first, the importance of citation verification, and the current limitations of these systems (like random regressions after model updates). For those wary of sending proprietary data to third-party services, the talk briefly explores running everything on your own infrastructure with open source LLMs. #AIAgents #LLMs #OpenSourceAI #ToolUse #SoftwareAutomation #LocalInference