This podcast explores how AI agents are transforming the work of senior engineers and technical leaders, moving beyond simple code generation to tackle complex infrastructure and architecture problems. The conversation demystifies the concept of evals, presenting them as a modern, quantifiable version of a product requirements document that allows models to determine the 'how' while engineers focus on the 'what'. The discussion also provides a practical framework for deciding which tasks to delegate to AI agents and how to capture subjective human taste into repeatable, scalable evaluation systems.