Ahead of the Field

AI agents in business: why the demo is the easy part and production is the hard part. Chris and Dan work through the 80% problem, the gap between an AI demo that looks right and an agent that does real work reliably inside a business, and what it takes to get agents working beyond ChatGPT.

In this episode: what has changed since ChatGPT launched three years ago and what is still marketing, what separates an AI agent from a chat interface, and how the cost of chat subscriptions compares with the API. They go through where agents are working in production today, including document processing, compliance onboarding, ID verification, invoice handling and reporting. They then cover what keeps those agents working: evaluations, auditability, and operating the agent as models and inputs change. They close on what changes tomorrow, starting with deciding where the real value is before building anything.

0:00 Moving Beyond the 80% Problem
1:36 What Truly Defines an Agentic System?
4:14 The Economics of AI Consumption
5:52 Enterprise Grade: Repeatability and Accuracy
13:33 The Bottleneck: From Prototype to Production
22:02 The Forgotten Pillar: Operate