LLMs vs. Agents vs. Harnesses: How AI Actually Works

In this episode of Why Isn't This AI Thing Working?, host Jess Lampe sits down with Alex Kozlowski, Director of Software Engineering at Launch Consulting, to cut through the jargon and explain how AI systems actually work.

Alex and Jess demystify the vocabulary everyone uses but few define: what a large language model really is, why LLMs are probabilistic (and why that makes hallucinations inevitable), and what separates an LLM from an "agent" and an emerging "harness" like Claude. Using plain-language analogies, they explain how tool-calling loops create the emergent behavior people find so impressive — and where these systems predictably go wrong.

They also tackle adoption reality: why teams with strong DORA metrics get the biggest gains, and why most AI proofs-of-concept never reach production.