| Decision | Episode | Date; publisher-listed duration | Why this belongs in the buyer’s queue |
|---|---|---|---|
| How do we evaluate before changing a model? | Ankur Goyal: Production AI Engineering starts with Evals | 11 October 2024; 117 minutes | Foundational evaluation and product iteration discussion |
| What changes between a checkpoint and a reliable service? | Philip Kiely and Ali Taha: The Inference Engineering Masterclass | 3 August 2026; 101 minutes | Serving configuration, caching, precision and workload trade-offs |
| Can a narrower model/API shape the product differently? | Diogo Almeida: Jev — System One models for Prod, not God | 21 September 2026; 141 minutes | Task decomposition, calibration and service primitives |
| Why does idle capacity change economics? | Erik Bernhardsson: Why AI Compute Comes in Rectangles | 22 October 2026; 86 minutes | Variable demand versus fixed compute commitments |
| What do we actually buy in a compute contract? | Evan Conrad: The Programmable Data Center | 20 October 2026; 37 minutes | Delivery constraints behind apparently interchangeable GPU hours |
The dates/durations come from the linked publisher records below. Short listening routes suggested here are editorial recommendations, not new measured results.