AI agents as microservices, but more

Why this conversation

Teams that learned to run microservices on Kubernetes now have to run AI agents there too, and the open source projects for it are moving fast. At Open Source Summit Europe in Prague, the host sat down with Lin Sun of Solo.io, who works on several of them and has been part of the Kubernetes community since its early days. The story of those three days in Prague is in Open Source Summit Europe 2026: who we met.

Ten years of KubeCon

The host opened with an observation: every time they open LinkedIn, Lin is in a different part of the world. Lin feels fortunate to speak in many countries and seasons, with Amsterdam twice that year, in March and again in September.

Lin’s involvement with the Linux Foundation runs through cloud native. Contributions to CNCF projects started about ten years ago, when Kubernetes had just started: the first contribution to Kubernetes was in January 2016. Lin has been a KubeCon speaker since 2017 in Berlin, and has attended and spoken at every single KubeCon in Europe and North America since. At KubeCon Amsterdam, Lin was one of the conference co-chairs, “so I was able to do a fun keynote for everybody.”

Agents as microservices, but more

What Lin focuses on now is the infrastructure that lets AI agents deploy, scale and stay secure in production, with the right governance and observability. One of the projects is agentgateway, which helps with the communication from AI agents to MCP servers, to large language models and to other agents, so that it can be secured, observed and governed.

“So you can think about AI agents as microservices, but more, right?” In the microservice world, the problem to solve was connectivity: how to connect, secure and observe your microservices. With AI agents the problem is the same, with a lot more focus on security, identity and policies, “because we don’t trust the AI agents”, and because agents can act on behalf of a person or a team. “That’s why a sandbox environment for AI agents is also becoming super important.”

Kubernetes for agents

Lin is excited about agentgateway and about kagent, which makes it easy to deploy AI agents on Kubernetes. A newer project, Agent Substrate, adds a layer of scheduling on top of Kubernetes, so that AI agents can be placed in a cluster on top of pods at efficient and large scale. Lin has started contributing to that one as well.

Half the time learning

The host listed the new protocols arriving for agents, MCP, WebMCP and CLIs, alongside talk of composability, accountability and evaluators, and asked how anyone keeps up. “It is hard. It is super hard.” Lin estimates that around 50% of the job is learning, because the space is evolving so fast. One good chunk of time goes to keeping up, another to building on top of these technologies, and another to fixing the problems Lin runs into, directly in the projects. That way, in Lin’s words, “I can not only be a consumer of these open source projects, but also be a contributor, be someone to help make the project successful.”

Where to join

Each project has a community meeting. agentgateway meets every Wednesday, and its website, agentgateway.dev, has the details; kagent.dev has its community information, with a meeting Lin believes is every Tuesday; Agent Substrate meets every Thursday. Agent Substrate is probably the largest community Lin has ever taken part in, with 50 people or more every time Lin joins. As Lin sees it, everybody wants to help shape the future of running AI agents in production on Kubernetes.

When it comes out

The episode is edited and in the release order. Like every episode it goes out on YouTube first, then on Apple Podcasts, Spotify and Amazon Music.