Brendan Burns, Kubernetes co-founder and Azure container infrastructure lead, shares his views on how AI is reshaping Kubernetes and software development. He explains how GPU scheduling, gang scheduling, and checkpoint-sensitive training workloads are forcing Kubernetes to evolve beyond its stateless origins. On AI app monitoring, he argues teams need human-in-the-loop feedback and large-scale prompt testing rather than simple error rates. Most provocatively, Burns contends that reviewing AI-generated code is a transitional phase — just as nobody reviews compiler output today, developers will eventually stop reviewing AI-generated code once testing frameworks and formal specifications mature. He also raises the question of whether future programming languages will be designed for machines rather than humans.
Table of contents
GPU scheduling is more than “another resource type”AI monitoring has to be differentCode review is just a phaseSort: