What I Took from KubeCon

Recently I joined KubeCon + CloudNativeCon Europe 2026 in Amsterdam.

Firstly, this is the biggest event I have joined in my life. I was shocked by how big it can be. I even got lost in the event center a few times. In total I did 40k steps.

I would not say I benefited the most from the event as a newbie. On the contrary, I would say I was bad at choosing which talk to join. I was even bad at reaching the talks. I missed one observability talk from Grafana. When I arrived there, it was already full and I was 5 minutes late.

Even though I suck at it, I think I got a few topics that I can share.

Why This Topic Stood Out to Me

The biggest topic was obviously AI and agentic development. Of course there is hype. But I think I got a few things that are important. Especially seeing changes from across the world. How everyone started to work differently. Also I started to think more about the future. It was not that a few prompts changed my life, but instead a few ideas, different point of views changed how I think about it.

What I Learned About Agentic Development

Firstly, yes, everyone relies on AI now for development. As a developer, our job is already getting affected by it. This is not the hype. I am barely touching code by hand anymore. It is more like giving directions, checking, reviewing, and testing.

Instead of getting defensive about agentic development, I should think about it more generally. Generally it is better to have speed with reasonable accuracy, if we can guide the agents in the right way and restrict them with least privileges.

Why MCP Servers Felt Important

We have MCP servers now but almost none of them have contributors, they are new tools so they are not developed enough to use but at the same time they are standardized for agents.

Personally if I deploy an app I don't want to check Argo CD first to see if everything is green, I can tell Claude Code to check. One of the best ways to do this is having a good MCP so the agent can understand its capabilities and turn natural language into machine-to-machine language.

Why RBAC Matters More Than Ever

This actually opens another important subject for everyone. Autonomous deployment, having a sandbox that you can define for your agent so it can also test what it wrote, importance of having RBAC on applications. Giving least privilege to agents.

This actually should be done even without agentic flow, but in general it is hard to apply. Having least privilege means making every app have role-based access and it is not supported by default. Now since we have agents doing everything and at the same time we don't know exactly what they will do, we can limit their access and let them work for us.

What I Want to Focus on Next

Right now I am trying to focus on limiting the applications I use. I am trying to create a custom wrapper that will have certain functionalities mostly read so I can give full permission to use my own wrapper. This is just a temporary solution until having full RBAC capabilities per app.

There is also the option to use MCP gateways until applications have all the capabilities. But so far MCPs are new and generally do not have all the capabilities yet. I would like to experiment more with the differences between CLI tools and MCP servers. If possible it would be better to use a universal protocol, so I prefer MCP, but we will see.