
Ankur Goyal, founder of Braintrust, argues that evals are the modern version of a PRD and that coding agents can run week-long benchmark experiments across database indexes and execution engines. He explains how to encode taste into repeatable evals and why fixing CI is the highest-leverage way to speed up engineering velocity.
Tap to vote and see what everyone thinks.
Summary by ByteBrief
The Hidden Complexity Cliff Inside Microsoft Power Apps