API Evangelist Conversation with Lin Sun on Gateways for the Agentic AI Era, Open Source, and Donating Projects to the CNCF

I’ve just celebrated being in the tech industry for twenty-five years. I started my career at IBM, and I almost thought I would end my career there. But five years ago I joined a startup, Solo.io, which was focused on the API gateway space back then, and we were in the thick of trying to gain momentum in service mesh. I brought my Istio background to the company, which is why I was hired. It was a big career change — going from one of the biggest tech companies to one of the smallest startups; it was about thirty people when I joined. My role at Solo.io is Head of Open Source. Over the past few years you’ve probably seen a lot of our open source work — we donated Gloo to the CNCF as kgateway, we launched kagent, and we launched Agent Gateway. I also just finished my term on the CNCF Technical Oversight Committee, and I’m a KubeCon co-chair for Europe and North America.

I actually really love giving talks. Some of the feedback I get is that I look like a different person on stage — I just get very excited, and I feel accomplished. I love that I can connect with my audience. Since we work in front of computers so much day to day, giving a talk is the best part of my job. In the past year I got really good at it, and that opens doors. After a talk at API Days in Paris, an organizer of Mobile World Congress invited me to speak there, no call-for-papers process needed. As a speaker I tend to tell a story through a live demo, so I put a lot of effort in. I used to just reuse old slides because I was the expert in the domain — but then the feedback at QCon was that I spoke too fast and went too deep. That woke me up. It’s not about how much you know; it’s about how you relate to people and make yourself memorable.

When I think about AI, it needs the same characteristic — something in the middle to mediate traffic. AI workloads are microservices plus. You still have all the traditional problems: you want to control, secure, and observe your services. But there are additional characteristics for AI. You want multiplexing and federation for your MCP tools. Security requirements are higher, because you don’t trust AI-generated code, so sandboxing and external authentication and authorization become more critical. And there’s the context layer — the messages and prompts flowing between agents, the model, and the MCP server — which we didn’t have to worry about with microservices, but now matters for governance and security decisions. I don’t think people will be all microservices or all AI workloads, so that puts pressure on gateways: can we have a unified gateway that handles both? That’s the role a project like Agent Gateway can play — mediating traffic to do policy, security, and governance, and providing observability for everything passing through.

AI is changing the dynamics. One of the foundations I’ve started participating in is the new Agentic AI Foundation, and some gateway projects are looking for a new home there, like Agent Gateway. The open source community is interesting because a lot of enterprises want an open architecture built in the open. If you’re running a gateway in Kubernetes — and a lot of people run gateways in production in the cloud on Kubernetes — there’s the Kubernetes Gateway API standard. But large enterprises aren’t only asking whether the gateway endorses the open standards; they’re asking whether the gateway itself is open source, and even whether the specific feature they need is in open source. They believe in adopting open, and they want to contribute as part of their journey. With everything changing so fast with AI, doing it in the open lets us iterate faster and leverage the community for testing and contributions, so the feedback loop is a lot faster.

You’ve probably seen how the MCP spec has evolved — a release candidate just went out, with the new release coming in July, and it has breaking changes. One of the things an API gateway can do, particularly one that provides MCP gateway functionality, is help users bridge spec changes rapidly. If the gateway can perform some conversion from the old spec to the new spec, it can provide a much smoother transition experience for MCP users. The tooling’s relationship with the spec is critical. The pace right now is extraordinary — progressive disclosure with MCP, code mode with MCP, how do we make MCP more efficient. Reading the changes in the new spec is genuinely exciting to me, because the technology keeps shifting and the gateway is where you can absorb a lot of that churn so your users don’t have to.

First I’d ask whether you’ve decided you want to donate to the CNCF and not another foundation — the Agentic AI Foundation might be worth looking at to see which is a better fit. But one thing I can share is that donating to the CNCF is a lot simpler now. There are three stages. The first is sandbox, where the bar is very low because we want projects that can innovate fast at the CNCF — the requirements are low and the process is fast. Once a project is submitted, the TOC normally reviews it within a month or two. The second and third stages, incubation and graduation, are where it gets hard and takes a long time. I tried to take a project to sandbox three years ago and it failed because the process was so complex — but now it’s much easier. They review fast and give you a yes or no, and if it’s no, they explain why and what you can do to make it acceptable.

I’ve been thinking about this every single week. After twenty-five years I challenge myself: if I didn’t get paid, would I do my job tomorrow? My job is very flexible. I love to speak, and getting to connect with the community and travel to a new place is super exciting. But the most notable factor is that I spend a huge percentage of my time learning new things with agentic AI. Every time I come back from a conference I realize the technology has changed. Imagine getting paid to learn new things — when you go to college you pay big tuition to learn, but now I get paid to do it, and that’s a good chunk of my week. Another chunk is speaking and connecting with people, and I also love writing content. Sometimes something inspires me and I want to write it down, and at that point it no longer feels like a job.