As a longtime cloud-native community member, I’ve been lucky to attend almost every KubeCon since 2016. KubeCon EU 2025 was the largest—and honestly, the best—KubeCon I’ve ever been to. Perfect weather, connecting with brilliant technologists, and consuming international eats in London—what’s not to love?
Maintainer Summit
Jet lag is always tough when traveling from the US to Europe, so I arrived Saturday morning and spent the weekend exploring London with Izzy. Monday kicked off with the Maintainer Summit, where the CNCF TOC shared updates on what they do, upcoming TAG restructuring, a leadership refresh, and how the community can get involved.
What I love about the Maintainer Summit is that it’s relatively small—most attendees are maintainers of CNCF projects, and the vibe is all about learning and collaborating on shared challenges. I joined a couple of great sessions, including an unconference discussion on evolving Kubernetes for AI/ML. These unconference formats are fantastic—attendees vote on the topics they want to discuss. (Check out the unconference notes!)
The Istio contributors gathered for a project meeting where we discussed future roadmap items like ext_proc and rate-limiting support for waypoints. Time flew—we barely scratched the surface of even one roadmap topic!
Co-Located Events
With it’s completely packed room, ArgoCon was likely the largest co-located event at KubeCon Europe this year. It was also my first time giving a full talk there, and I challenged myself with a mostly live demo—showing off a generative AI app I built with RAG services. Using Argo Rollouts, I deployed version 2 of the RAG service, which added support for uploading both PDF and TXT files into a vector database.
The demo ran in Istio Ambient, secured with mTLS, and used the Gateway API (Gateway and HTTPRoute) to control traffic into, out of, and within the cluster.
My demo failed in the final few minutes, and I didn’t have time to debug live—so I resorted to the backup recording 😅 (blaming the notorious conference Wi-Fi!). Catch the YouTube video below.
Sadly, I missed most of the IstioDay talks due to a TOC offsite, but I caught the last few—very interesting sessions. Denis Jannot from Solo.io showed a live load test of Istio’s various deployment modes (no mesh, sidecars, Ambient L4, L4+L7), using scripts from a repo linked via QR code.
Gerald Kleser and Darren Hague from SAP gave a great overview of their GenAI platform, which was built on Istio, Kubernetes, Argo Workflows, KServe, and vLLM. I was surprised (and thrilled!) to see SAP already adopting Istio ambient and recommending it for mTLS use.
Rob Salmond also delivered an awesome talk on how to get Istio help—his presentation style is top-notch!
Service Mesh
There was tons of excitement around Istio ambient mesh this year. After 8 years working on service mesh, I feel like we’ve finally reached a point where it’s truly transparent to application workloads—just as we envisioned at Istio’s inception.
I’m now confident enough to pull off an Istio ambient demo on the fly—it’s simple to show and pretty easy to troubleshoot. Here are a few highlights:
- Faseela K and I read our book Izzy Saves Birthday and did a live demo of Istio Ambient securing, controlling, and observing traffic.
- Stephen Connolly from HSBC shared how his team worked with Solo.io to adopt Ambient and significantly cut costs.
- Ahmed Bebars and I presented The New York Times’ journey into Ambient, complete with a dramatic live demo featuring a classic “upstream connection error” and recovery—one of my favorite talks ever! We spoke entirely without notes and connected deeply with the audience through live storytelling.
- Raymond Wong from Forbes joined Louis Ryan and me to share lessons learned from adopting Istio Ambient in production—huge thanks to Forbes for sharing their story!
- Louis Ryan, CTO of Solo.io and co-founder of Istio also shared a few exciting roadmap items for Istio, including Windows support, the Gateway API Inference Extension support, multicluster Ambient, and more.
If you’re considering migrating from Istio sidecars to Ambient, Idit Levine, CEO and Founder of Solo.io and Keith Babo, Chief Product Officer at Solo.io announced a free cost-saving estimator and migration tool built by the Solo team that is free for everyone to use during Thursday’s keynote. Check them out:
AI everywhere
First, AI workloads bring new challenges:
- GPUs are expensive, and availability varies—how do we intelligently route requests to ready endpoints?
- How should AI workloads be packaged in a standard way?
- How can we effectively perform health checks when response times vary widely?
- How can we make Kubernetes run LLM effectively? Why does running LLMs like Ollama on Kubernetes feel slower than on local machines?
To learn more about the first challenge and how the Kubernetes community is solving it using the Gateway API inference extension, check out Daneyon Hansen and Abdel Sghiouar’s talk.
Second, how do we leverage AI to solve some of our cloud native operation challenges given the complexity of the cloud native projects and our growing landscape?
Enter Kagent, which was open-sourced just three weeks ago. It’s already generating a lot of interest and is being donated to CNCF as a sandbox project. Kagent was spotlighted as one of the future AI projects in an invitation-only CNCF strategy session. We’re seeing contributors outside of Solo.io jumping in, and early adopters are already deploying it.
I really wish I had used Kagent in my live demo—it’s now on my to-do list.
In the new Demo Theater event at KubeCon, Christian Posta gave an awesome demo of Kagent and its tools for tackling cloud-native ops challenges.
The Modern Gateway
If you use Kubernetes, you know how crucial gateways are for controlling, securing, and observing traffic in and out of your cluster.
This year’s KubeCon featured dozens of talks about gateways—both from project maintainers and users. Gateway APIs have matured past the shiny-new-thing phase for standardization around gateways; people are actively figuring out how to adopt them in production.
With agentic AI workloads rising—and Kubernetes as the preferred platform—gateways are becoming even more critical. These AI agents rely on tools to accomplish tasks and communicate with LLMs, often using the Model Context Protocol (MCP), which acts like a USB-C port for AI applications, standardizing how models connect to data and tools.
Kgateway, a CNCF sandbox project, has implemented the AI Inference Extension to handle smart routing for LLMs. We shared our community roadmap during in the latest kgateway update blog which focused on two highlights:
- AI Gateway feature (including the Gateway API inference extension implementation)
- Kgateway as a waypoint for Istio Ambient
With the release of kgateway 2.0, I’m thrilled to see these features land—huge shoutouts to all of our maintainers and contributors for their hard work! Release notes here.
As innovation continues, we’ve also built an MCP Gateway inside kgateway. It lets you control, secure, and observe traffic between MCP clients and servers—super useful as the ecosystem grows more complex. With the rising popularity of MCP and the sprawl of MCP servers and tools, having a gateway to mediate the traffic makes a lot of sense.
Post-KubeCon Excitement
I’m still absorbing everything I learned at KubeCon—following social media, reading blogs, and experimenting hands-on with MCP Gateway, kagent, and kgateway for Istio Ambient. Can’t wait to catch up with some of the talk recordings as they come out!
As someone building GenAI apps (and using Cursor + ChatGPT daily!), I know AI is something we can’t ignore. It’s an exciting time to be at the center of this innovation, and I’m incredibly grateful to be part of the cloud-native + AI journey.
You can catch up with all the latest announcements from the Solo.io team at KubeCon + CloudNativeCon Europe 2025 on our new community page here.
.png)
