Four conversations from KubeCon Europe 2026

Four episodes of the show were recorded in one week at the RAI in Amsterdam, during KubeCon + CloudNativeCon Europe 2026. The show was there on a CNCF press badge, and the host, Luca Berton, was also on the programme as a speaker. The overview of where the show films gives that week a few lines. This is the longer account: what each conversation covered, what ties them together, what else the host saw, and where to read more.

Security starts with what you already run

Kat Cosgrove leads developer advocacy at Minimus. She also holds an elected seat on the Kubernetes Steering Committee, leads SIG Docs and owns the Release Team subproject. She did not start with a product. She started with Ingress NGINX, archived on the KubeCon stage the day before, while a little under half of cloud native environments still ran it. Her argument: the open source a company depends on in production is part of its security posture, and so is keeping it maintained. The project had warned its users for a long time. Staying on it now is, in her words, “a deliberate choice to put yourself in danger.”

The second half moved to the container image. Fewer packages mean fewer CVEs, and an image has to be rebuilt every time upstream ships a patch, not when someone remembers. Signed SBOMs, FIPS and STIG images and air-gapped registries came up as well. Her closing line set the bar for the rest of the week: “we don’t want security to be a thing that’s hard for people to implement correctly.”

The host’s written interview with Kat Cosgrove is on lucaberton.com.

Observability you are free to leave

Sven Mößbauer of Dash0 talked about the cost of staying. Many vendors on the floor said they were OpenTelemetry native. His point was that much of that data is turned back into a proprietary format once it arrives. He named two kinds of lock-in. The technical kind: data, queries, alerts and dashboards that only make sense inside one product. The commercial kind: contracts of one or three years while telemetry volumes keep growing. Dash0’s answer, as he described it, is open standards throughout, with OpenTelemetry, PromQL, Perses for dashboards and Prometheus for alerts, and a consumption model instead of a contract.

On AI he was exact: “AI is only as good as the context that you provide it with.” An assistant added on top of a platform that does not understand its telemetry gives vague answers.

The host’s written interview with Sven covers the same ground in text.

Platforms for code that agents write

Luca Galante works on the Platform Engineering community: PlatformCon, its World Tour, House of Kube and the research arm, Weave Intelligence. That week he had presented research on agentic coding in the enterprise, which sorts teams into levels of agent adoption. His claim runs against the expected one. AI does not make platform teams redundant: “AI is actually turbocharging the need for well-designed platforms.” A prototype now takes hours. Getting it to production, and keeping it running long after launch, still takes golden paths and guardrails. When every developer’s agents feed the same pipelines, the bottleneck moves downstream. The internal developer platform has to grow into what he calls an agentic development platform, and platform and data teams have to work from one table.

The host’s written interview with Luca Galante has the full outline.

A European public cloud

Guido Laout is Director Netherlands at OVHcloud. His conversation was recorded on Thursday 26 March, the last day of the conference. The episode is edited but not yet published, and it has no release date. Its page will carry the full notes once it is out. Until then, the host’s written interview with Guido Laout records what reached the OVHcloud stand that week: steady interest in sovereignty, questions about the managed services, and why an organisation should have an exit plan.

The thread

Four guests, one direction: each moved a hard problem off the individual engineer and onto the platform. Kat Cosgrove wants the secure path to be the default one. Sven wants telemetry that can leave with you. Luca Galante wants guardrails for code produced faster than people can review it. Guido Laout’s subject is where the platform runs, and under whose rules. None of them described a launch. Each described an obligation that lands on the team that runs the platform, and two of them said AI makes it heavier, not lighter.

What else the host saw that week

The host’s notes on lucaberton.com fill in the rest of the week.

• Monday, the co-located day. Platform Engineering Day, Observability Day and ArgoCon Europe, written up slide by slide, including an Argo CD tuning talk with before-and-after figures. In the evening, Signal Overflow, the SRE NL meetup at Booking.com, where the host and his co-author Grzegorz Stencel signed copies of their book, Kubernetes Recipes.
• Tuesday, the talk. A second signing at the vCluster booth, then the host’s own session: an experience report on sharing NVIDIA H200 GPUs between teams on bare-metal OpenShift AI, managed through GitOps. The recap says 780 people had saved the session in the schedule, and links the slides.
• All week, the hallway. Fifteen short clips with people met in corridors and at booths: first-time attendees, contributors, one of the event’s co-chairs, observability vendors on the cost of telemetry, edge Kubernetes demos and a cluster in a cat carrier.

Where to read more

• The week in photos, day by day.
• The co-located day, talk by talk.
• The hallway interviews, fifteen of them.
• The talk recap, with the slide deck.

Next

New episodes every Tuesday, on YouTube first, then Apple Podcasts, Spotify and Amazon Music; every platform is listed on the home page.

Built something the rest of us run on? Being a guest is free: how it works.