You Probably Don't Need an AI Agent | ft. Chris Colinsky @Howl ( ex-@goop ) | #TechUnhinged

In this episode, the CTO and principal AI engineer at Howl, and former CTO of goop and MeUndies, explains why most companies chasing "AI agents" actually just need a well-engineered workflow with an LLM baked into it. Chris Colinsky draws on 20+ years building for the web since 2003, shares how he finished near the top of OpenAI's parameter-golf leaderboard without any prior model-architecture experience, and breaks down why AI mostly amplifies whatever process, good or bad, you already have. We also get into why most AI failures are really data problems in disguise, his "calculate first, reason second" method for building reliably on unpredictable models, graph-based retrieval over traditional RAG, self-hosting small models for cost and privacy, and his sharpest line of the conversation where he says, AI didn't create your data problem; it just removed your ability to ignore it. 00:00 Introduction 00:41 What Is An LLM 02:08 LLM Limits And Misconceptions 03:47 What's New In AI Engineering 07:06 Modernizing Legacy Systems 10:19 Why AI Fails: Bad Data 14:05 Calculate First, Reason Second 15:26 Graph RAG And Data Layers 19:03 Building The Open Armature Framework 25:23 Leadership, Adoption And Getting Started Subscribe for more episodes of ‪@TechUnhingedPodcast‬ Catch more episodes here: Website URL: https://techunhinged.com/ LinkedIn:   / techunhinged   Instagram:   / techunhingedpodcast   Soundcloud:   / techunhinged   Powered by ‪@codedistrictofficial‬​