Guest: Rob Hirschfeld (LinkedIn)
Company: RackN
Show Name: KubeStruck
Topic: Kubernetes

KubeCon has spent a decade shaping the cloud-native movement, but this year’s event revealed something unexpected. While AI dominated almost every conversation, several critical infrastructure topics quietly vanished from the agenda. Few people have the perspective to read these shifts clearly — but Rob Hirschfeld, CEO and Co-Founder of RackN, has been attending KubeCon since it fit inside a San Francisco hotel. His long view reveals what changed, what normalized, and what gaps now threaten to slow Kubernetes adoption in the enterprise.

Rob Hirschfeld has watched KubeCon evolve from a small community gathering to one of the most influential events in modern infrastructure. At this tenth-anniversary edition in Atlanta, he immediately noticed both the scale and the shift in tone. The show floor stretched across the entire convention center, packed with both legacy names and brand-new entrants. But what stood out most to him was not the size — it was what the event did not talk about.

He explained that certain topics that once dominated conversations have nearly disappeared. “There really are waves of things that we see,” he said. A few years ago, internal developer portals (IDPs), Terraform orchestration tools, and the entire “tacos” ecosystem were everywhere. This year, they were almost entirely absent. The same happened to the once-heated observability debates. Instead of competing pitches, observability has now become an assumed, foundational layer — a normalized part of the stack.

GitOps also moved into that same category. What once required extensive explanation and evangelism is now simply understood as the default way teams operate. Continuous integration platforms, pipelines, and automation tools have matured to the point where they no longer feel like cutting-edge innovations. They are expected.

But while some topics have normalized and others have faded, what concerned Hirschfeld most this year was a glaring gap: the almost complete absence of infrastructure-level Kubernetes discussions.

RackN specializes in bare-metal automation, making Hirschfeld particularly attuned to what’s happening — and not happening — in the infrastructure space. “There’s zero talk on bare metal,” he noted. “There’s zero talk on Cluster API. There are only a handful of talks about kubevirt, even though enterprises desperately need a VMware replacement.” For him, the lack of content around these operational foundations was striking.

Ten years ago, Kubernetes sessions revolved around installing clusters, building control planes, and understanding the low-level mechanics. Even five years ago, infrastructure topics still had space in the conversation. Today, those topics have been pushed aside, overshadowed by higher-level concerns or assumed to be solved problems.

That assumption, Hirschfeld argued, is dangerous.

Enterprises, especially those dealing with repatriation or hybrid environments, still need guidance on how to run Kubernetes on bare metal, how to handle virtualization transitions, and how to manage infrastructure for AI workloads. These gaps are not theoretical. They directly impact teams attempting to modernize their environments or migrate away from legacy virtualization solutions like VMware.

At the same time, the community’s attention has shifted toward AI — but even that focus is more surface-level than operational. According to Hirschfeld, most AI discussions at the event were about running workloads on Kubernetes or integrating AI analyses into tools. But deeper topics — such as building AI clusters, managing bare-metal GPU infrastructure, or optimizing environments for training — barely appeared.

The audience itself looked different this year. Hirschfeld noted that the attendees he encountered were primarily SREs and developers looking to run workloads, not infrastructure teams trying to design foundational systems. This shift in audience composition also influences what makes it onto the agenda. “We’re still very much about the functional unit of work inside the Kubernetes ecosystem,” he said. Load balancers, security controls, pipelines — the components that sit on top of the platform — dominated the conversations.

This left a disconnect between what enterprise buyers need and what the community platforms at KubeCon explicitly talk about. The enterprises Hirschfeld works with care deeply about infrastructure automation, virtualization replacement, and large-scale cluster design. Yet these needs were not reflected in the content selected for the show.

The disconnect isn’t intentional — it’s a result of how big and complex the ecosystem has become. “The show sells out fast, and sessions are booked very early,” Hirschfeld pointed out. By the time market shifts happen, the agenda is often already locked. This lag can lead to events that don’t fully represent the industry’s current pressure points.

The other major force reshaping Kubernetes is AI. While Hirschfeld expected AI to take over the show, it wasn’t as dominant as he assumed. AI was present in almost every conversation, but it wasn’t the overwhelming headline theme. More importantly, AI is beginning to change developer workflows in ways that reduce the need for tooling once considered essential.

A few years ago, developer portals were a solution to the complexity of YAML files and CI configurations. Today, large language models can generate those files instantly. Developers can stay inside their IDEs, ask for a configuration, and get a working example. The argument that developers need portals to avoid learning Kubernetes manifests is no longer as strong.

This introduces a new problem: developers may not actually know if the AI-generated configurations are correct. “They don’t always know if AI is doing a good job,” Hirschfeld warned. This emerging uncertainty creates risk, and the community has not yet built best practices or guardrails around AI-generated operational artifacts.

The conversation also turned reflective. When asked whether Rob—who was in the room when Kubernetes was first announced as an anchor project—ever envisioned this future. Hirschfeld admitted he did not. He recalled early debates about containers versus virtual machines. “There was a lot of belief that containers would wipe out VMs,” he said. Instead, Kubernetes became the virtualization control plane itself.

Even Hirschfeld was surprised by the way the virtualization market has evolved. Kubernetes did not replace VMs — it became the system that orchestrates them. Meanwhile, traditional virtualization vendors gradually aligned themselves with this new reality.

Another long-term surprise is the continued dominance of the CLI. Even after ten years, kubectl remains the primary interface for most users. Despite expectations of sophisticated UIs and management consoles, the command line has persisted. Hirschfeld pointed out that someone could probably run the exact same demo they ran ten years ago using kubectl, and it would still feel natural today.

As the conversation wrapped up, Hirschfeld highlighted a consistent challenge: the Kubernetes ecosystem keeps growing, but the need to integrate countless point solutions makes the landscape overwhelming. Enterprises benefit when a curated platform abstracts complexity, which explains why many buyers gravitate toward large vendors like Red Hat or the cloud providers.

However, the growth of AI, the pressure to replace legacy virtualization, and the lack of infrastructure-focused sessions leave clear signals for the future. The operational side of Kubernetes — once central to the community — is slipping into the background. Whether that gap becomes a risk or a sign of maturity depends on how the ecosystem responds.