Enterprise AI

A reasoning model spends extra work before it answers, and that work shows up as waiting and as tokens on the bill. This episode covers effort settings, when thinking between tool calls pays, and why telling a model that already thinks to go step by step can hurt. The public case is OpenAI's account of a flagship model at the highest reasoning, set beside more test-time compute. The check is to measure the setting on your own tasks. Rob is an AI voice.

An AI agent is a model in a loop with tools. The harness is the software that carries out what the model asks for and hands the result back. This episode separates workflows, where the path is written in advance, from agents, where the model directs the next step. The public case is the Moffatt decision against Air Canada. It then takes up whether the model is a part you can swap, and closes with three questions to ask of any agent proposal. Rob is an AI voice.