.@braintrust is building the infrastructure for production AI. We're very excited to be backing their Series B. CEO Ankur Goyal joined a16z's Martin Casado to cover how systems engineering stacks up against the bitter lesson, why Chinese models are gaining ground faster than spending suggests, whether AI demand will eventually saturate, and more. 00:00 Introduction 03:10 What "evals" actually mean (hypotheses and the scientific method) 07:34 AI is continuous, systems are discrete 08:29 The bitter lesson: universal function approximators vs. engineering 11:58 Engineering the harness, not the model 17:47 Chinese models: high token usage, low dollar share 20:37 Why open source models aren't dominating yet 27:08 Frontier labs, infinite capital, and scaling limits 30:38 Demand-side saturation: enterprises can't absorb better models 38:47 Bash vs. SQL for agents: "the results are comical" @ankrgyl @martin_casado(a16z)