Events

What we’re taking home from KubeCon NA 2026

We came to Salt Lake City to spend time with the cloud-native community. Our week took us from hundreds of conversations at booth 340 to three evenings together around the city.

Russel Estardo

Four Rootly team members together at the KubeCon booth in front of the blue and purple Rootly AI backdrop

For four days in November, thousands of engineers, maintainers, and open-source contributors gathered in Salt Lake City for KubeCon + CloudNativeCon North America. Our Rootly team spent the week at booth 340, having hundreds of conversations with the people building and running cloud-native software.

In the evenings, we traded the expo floor for tables at Ivy & Varley, Gracie’s, and Poplar Street Pub. Across three happy hours, there was time to share a meal, get a game of pool going, and settle in with people we might otherwise only have met in passing.

Four themes from the week’s program and early conference reporting stood out for our work on reliability.

Guests seated together over food and drinks at the bar
Monday evening at Ivy & Varley, before the conference schedule took over.

AI infrastructure is now an on-call concern

This year’s dedicated AI Inference + Agentic track put model serving, GPU scheduling, and production observability alongside the rest of cloud-native operations. In his opening-keynote recap, Bill Doerrfeld picked out the pressure on resources and the need to enforce boundaries around agents.

Operational questions belong at the start of an AI rollout. Who gets paged when inference slows down? Which customer journeys depend on that model endpoint? What can the service do when capacity runs out? Those answers need owners before the workload becomes critical.

For reliability teams, that also means looking beyond whether a model answered correctly. Latency, queues, dependency failures, and degraded behavior belong in the same conversation.

Better incident context is part of the AI work

The session The SRE Is Not Dead described a financial-services incident response effort hampered by missing deployment history, ownership, and dependency context. That is a useful counterweight to the idea that adding an agent automatically makes an investigation easier.

Make it easier to answer “what changed?” and “who knows this service?” A metric can point to a symptom. The surrounding history helps a responder decide where to look next, whether that responder is a person or an AI assistant.

This is also why the work after an incident matters. Decisions, timelines, and follow-up actions become context for the next investigation. We want incident AI to show the evidence behind its next suggested step.

A guest chatting across a wooden table at the afterparty
Continuing the week’s conversations at Gracie’s on Tuesday.

Automation needs a way back

The abstract for GitOps Meltdown described cascading deletion across a platform managing 39 clusters, including the Flux controller itself. It puts a sharp question to any automation project: how do you recover the system that normally performs the recovery?

The same question belongs in AI operations. Doerrfeld’s Agentics Day takeaways highlighted access controls, monitoring, and governance around the tools agents can use.

Before expanding an automated workflow, we want to know its scope, how someone stops it, and what remains available if it fails. Rehearse the rollback. Keep the record of what changed. Make approval meaningful by showing the proposed action and its likely impact.

The developer platform needs its own recovery plan

The workshop The Most Boring Talk at KubeCon focused on failures across the developer portal, version control, and delivery pipeline.

Include the platform in reliability planning: ownership, escalation, tested restores, and a way to communicate when the usual tooling is unavailable. Test what happens when credentials expire or the portal’s database needs restoring.

One useful exercise is to walk through a recent incident and ask which parts of the response depend on the same infrastructure that failed. Our unofficial SRE track collects more of the sessions that informed this perspective.

Away from the conference floor

We also made time to get everyone away from a screen. On Monday, we hosted our kickoff at Ivy & Varley with Twingate and RWX. Tuesday brought the afterparty at Gracie’s with ClickHouse, Mezmo, Tailscale, and Plakar. On Wednesday, we gathered at Poplar Street Pub with Chainguard, Zesty, Twingate, and Runpod.

The photos tell that part of the week well: shared food, a toast at the bar, a game of pool, and a seat beside the patio fire.

A guest lining up a shot at the pool tableTwo guests standing on opposite sides of the patio fire pit
A pool table and a patio fire at Wednesday’s gathering at Poplar Street Pub.

Thank you to everyone who stopped by our booth or joined us after hours, and to the partners who helped bring the gatherings together. The three recaps below collect photos from each evening.

Your privacy choices

Rootly uses cookies from advertising partners to measure our ads and to show ads for Rootly on other sites. Some US state laws call that selling or sharing personal information. You can opt out of it for this browser here.

Analytics we use to improve this site stay on. Read Rootly's Privacy policy for more information.