KubeCon 2026: Kelsey Hightower on AI and the Expertise Reset
Cloud Platform EngineeringKubeCon + CloudNativeCon Europe 2026 drew a record 13,500 attendees to Amsterdam. The people most drawn to open source infrastructure are the same people now having to figure out how AI changes what they build — and that convergence is part of what drove the attendance record.
Kelsey Hightower, who retired from Google in 2023 and remains one of the most cited voices in the Kubernetes community, used his KubeAuto Day keynote to offer a frame most AI commentary misses: “In a way, everyone is a junior engineer when it comes to AI.” Not as a dismissal, but as a leveling observation. The experience hierarchy that took decades to build in cloud-native infrastructure does not transfer automatically to a field that is, by any honest measure, genuinely new.
The Open Source Prerequisite
Hightower’s argument is less about AI capability than about the prerequisite for using it well. His contention: engineers who won’t contribute to open source and maintain it won’t be able to work effectively with AI systems either.
The logic is specific. Open source produces the training data, the documented patterns, and the production-battle-tested tooling that AI systems are trained on and operate within. “Who do you think trained the models we did? These are our ideas. They’re mimicking our creativity. So when I write the docs, and I write the code, those are my experiences being committed and serialized.”
The deprecation of the NGINX Ingress controller — a significant discussion thread at this year’s KubeCon — gave that argument a concrete edge: “Lots of enterprises are scrambling now, and not one of them that I’ve met has thought about maybe they fork it and step up to maintain it. I think a lot of people have to remind themselves that open source wasn’t about getting software for free from someone else. It’s also about stepping up to maintain it when the time comes.”
The AI era accelerates this dependency without automatically producing more contributors. Organizations consuming AI-assisted code generated from open source patterns, without contributing to the foundations those patterns rest on, are compounding a structural fragility they haven’t yet accounted for.
The Security Understanding Gap
Hightower’s second concern has the most direct operational implication. AI tools are removing engineers’ ability to understand their own systems at precisely the moment that understanding matters most.
“A lot of people have said they have lost the ability to understand if something is secure or not. What use are you at that point? This is the danger of any system that removes people’s understanding, because when it’s time to understand, you won’t be able to.”
This is not a generic “don’t rely on AI” argument. It’s a specific observation about how delegating understanding creates brittle expertise — functional in normal operating conditions, absent when something fails in a way the AI tool didn’t anticipate. His corollary on full-stack knowledge is direct: “If you’re a practitioner and it is your responsibility to make sure Kubernetes actually works, that is something I think you need to understand.”
His advice on AI tool evaluation follows the same empirical pattern: “Go find someone actually using it in production. Can I get 30 minutes with you? I have some honest questions. What breaks? Why should I not use this?”
The So What
The “everyone is junior on AI” framing is not pessimistic — Hightower’s KubeCon closing was characteristically optimistic. But it has a practical implication for platform and SRE teams: experience in Kubernetes, observability, or distributed systems does not automatically translate into senior judgment on AI infrastructure. The field is new enough that pattern libraries are thin, failure modes are not yet well-documented, and production intuition has to be rebuilt from near zero.
The engineers best positioned through this period are doing what they’ve always done: learning the patterns, taking notes, and becoming “the person who says, from experience, this is what works, this is what doesn’t work.” AI changes the content of that learning. It does not change the process.
Photo by Luis Quintero on Pexels
Source: The New Stack
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →