.jpg)
Earlier this year at KubeCon + CloudNativeCon Europe 2025 in London, podcast hosts Ronald Kers and Jan Stomphorst finally managed to corner our Principal Architect Kevin Reeuwijk.
The resulting Dutch‑language episode of De Nederlandse Kubernetes Podcast is a roller‑coaster tour of orchards full of robot pickers, factory lines that snitch on their own inefficiencies and the thorny business of moving thousands of virtual machines off vSphere.
If you understand Dutch, you can listen to the full recording here. Below is an English recap that digs a little deeper than the show notes — highlighting the technical meat, the practical lessons and, of course, Kevin’s best quotes.
Let’s get going.
“I actually wanted to be an air‑traffic controller,” Kevin began. “I know there are people who want to follow the book very strictly and never deviate from procedure — but that isn’t me. I like solving the problem in front of me, even if it means bending the rules a little. Apparently that was too creative for the control tower.”
That mindset has shaped Kevin’s career as he moved through the world of IT infrastructure, from Windows admin to data center CTO, and finally to declarative and cloud native with Puppet and now Spectro Cloud.
Edge Kubernetes puts clusters in the strangest places
And what’s he doing at Spectro Cloud? Edge is a big focus, and that’s where the interview really kicks off.
“I've been a Principal Architect at Spectro Cloud for the last three years. The world of Kubernetes is really exploding, and edge, where we’re very specialized, is becoming very popular — so we are busy!”
“The edge doesn’t look like the neat, well‑lit racks you have in a data center,” Kevin told the hosts. “You ship devices into networks you don’t control, sometimes with no one on‑site and no Cluster API. Everything still has to work when the lights go out.” Power cuts, holiday shutdowns, ever‑changing IP ranges — it all has to just work.
Kevin shared three examples of how edge manifests in the real world.
Tethered drones that never need batteries
In fields and orchards, Tevel deploys fruit‑picking drones, picking only perfectly ripe fruit, 24x7, through advanced computer vision models. “Instead of hiring seasonal workers… drones with computer vision look at an apple from all sides, see if it’s ripe, then a suction cup picks it.”
The compute horsepower — high‑end GPUs — lives in a rolling ground station. “If we put the GPU in the drone it would be far too heavy to keep airborne,” Kevin explained. Instead, each flyer trails a combined power‑and‑data cable.
The system shows how Kubernetes can orchestrate swarms of specialized hardware when latency is non-negotiable and batteries are precious.
Goodbye gag‑inducing dental molds
Switch scenes to the dentist. Remember biting into that rubbery tray and fighting the gag reflex? “I always used to gag a little when they pressed the mold into my mouth,” Kevin recalled.
Dental tech giant Dentsply Sirona’s wireless intra‑oral scanner is a much more elegant solution.
“A handheld camera scans the inside of your mouth… a smaller unit with a GPU processes all data and the doctor works from a tablet.”
The system renders a 3D model in real time on a compact NVIDIA‑powered box beside the chair. Only processed data leaves the clinic — handy when the router reboots during the holiday season. “The device might be powered off for weeks, and the DHCP lease for the IP address could well have expired, causing the edge node to come up with a new IP address. It's our responsibility to make sure that IP change doesn't break the cluster. That’s the kind of thing you need to prepare for at the edge” Kevin said.
A factory line that tattles on itself
Even cigarette packaging gets the edge treatment. “Cigarettes are not exactly healthy, but someone’s got to box them,” Jan joked. Spectro Cloud works with a machinery manufacturer that helps do just that. It now ships each high‑speed line with an embedded Kubernetes cluster that runs Kafka locally, streaming metrics for AI‑driven efficiency tweaks — even when the factory network is completely isolated.
Why disconnected beats centralized
To survive these conditions Kevin argues for autonomy: every device carries enough brains to reconfigure itself and phone home through firewalls. “Platforms like Rancher try to do everything centrally,” he said. “When the link goes down, you lose control. We embed an agent so the cluster keeps working even if the management plane disappears for weeks.”
That independence starts at install time. For high performance, virtualization or AI clusters, Kevin recommends bare‑metal Kubernetes bootstrapped via Canonical MaaS: PXE lays down Ubuntu or RHEL, an agent forms the cluster and “even if you turn off our management platform, the cluster can still run.”
Extending Cluster API ideas beyond nodes and networking, a cluster profile captures the whole stack — from OS image to application manifests — ready to stamp out ten clusters or ten thousand.
Escaping vSphere: what enterprises really need to replace
The conversation turned to the other topic on every platform team’s mind: legacy VMs. VMware’s catalogue is sprawling — “more than two‑hundred, maybe three‑hundred SKUs,” Kevin laughed. But when Broadcom announced price hikes, most customers worried just about the core vSphere capabilities: snapshotting, vMotion and DRS.
Spectro Cloud’s answer is a packaging of KubeVirt plus surrounding services, to tackle the core enterprise feature set. “That’s where we focused,” Kevin said. “Templates, snapshots, backups, live migration and mixed‑CPU scheduling… if you give people that, they can move. We call it VMO — KubeVirt with a whole mountain around it — to give you enough of what vSphere Enterprise does so you don’t feel you’re losing anything.”
KubeVirt labels every node with its CPU flags and chooses the largest common denominator so a VM can live‑migrate across generations. “Funny enough, it’s a bit better than VMware’s EVC in that regard, as EVC can't be changed while VMs are running, while Kubevirt can,” Kevin noted.
Of course, vMotion is only as good as the storage underneath. Enterprises juggle arrays from “supplier A, B, C,” Kevin said, “and Ceph can do it, but it’s super complex and resource‑hungry.” Portworx, Pure, Hitachi, Dell CSM, Piraeus and others already support RWX block modes in a much less resource-hungry way that makes them “perfect for the KubeVirt use case.”
Rather than pick a winner, Kevin’s team publishes reference architectures that prove cloning, snapshots and live migration across Ceph, Portworx, Pure, NetApp and more. Preserving choice is a key consideration when dealing with demanding enterprises — and a core belief at Spectro Cloud.
Will AI take your job?
Finishing up, the hosts asked Kevin his predictions for the future, and AI of course came up. The hosts couldn’t resist asking whether operators will automate themselves out of work.
“The answer is actually always yes,” shrugged Kevin. “What I expect a lot of is AI-based operators for managing Kubernetes clusters: all the things we do to keep Kubernetes clusters running, that we check to ensure that applications stay up. I hope we'll get to the point at some point that we can ship something like an AI operator on the platform so that keeping Kubernetes clusters running is something you don't have to think about anymore. “
Until then, being too creative sounds like a career advantage.
If that whets your appetite, dive into the full Dutch episode — it’s as entertaining as it is technical. And next time you bite into a crisp apple (picked by a tethered robot) or sit for a brace fitting (without the goo), you’ll know there’s a Kubernetes cluster working hard behind the scenes.