Vinay Seshadri joins Sophia Dew to describe Agent Usage Detail Records, an open data standard intended to join consumption across applications, agent harnessesAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction. and model routersAI model routing sends each request to a model chosen for that request's complexity, cost, speed, privacy, or other requirements.. He draws a parallel with telecom call records, where separate systems needed a shared format to explain the cost of an entire call.
The discussion distinguishes predictable seat-based software revenue from highly variable agent workloads. A customer's task mix, model choice and tool callsAI tool calling lets a model request a structured operation from software instead of only returning natural-language text. can change costs sharply between runs, making revenue easier to measure than customer-level margin. Seshadri argues that runtime instrumentationAI cost observability explains which AI runs, users, models, tools and behaviors generated resource spending. is needed before those costs can inform pricing.
Lower model pricesInference cost is the expense of running a trained AI model to process inputs and produce outputs. do not automatically remove harness, evaluation and third-party API costs. The interview explores routing decisions, adapters and feedback on the proposed standard, while treating the claimed cost swings as the guest's operational examples rather than a universal benchmark.
Watch on YouTube
