GitLab Transcend London: My recap from the stage.

Event Day: Waiting for your name to be called.

The next morning started off quietly. Breakfast, another cup of coffee, and then mostly just waiting. Job was already in the auditorium, while I was backstage until it was my turn to go on stage. Since the program ran a little late, the wait took longer than expected. At a moment like that, you can’t help but go over everything in your head one more time. Am I telling my story well? Am I forgetting anything? And then suddenly you get the signal, and it’s time.

To be honest, my nerves disappeared almost as soon as the conversation started. I was on stage with Ryan from Compare the Market, Matteo from AWS, and Gene Kim from IT Revolution, and it felt like the time flew by. During the session, we also announced the updated Cube × GitLab case study, which made the experience even more special. What I only found out afterwards might have been the highlight of the day for me. Back at our office in Oldenzaal, my colleagues had organized a viewing party, and apparently the room went completely quiet as soon as I started speaking. Just knowing that made the whole experience even more meaningful.

What stuck with me the most: context.

If I had to choose one theme from Transcend, it would be context. Virtually every presentation and discussion ultimately came back to the same idea: AI only becomes truly valuable when it understands the context in which it operates. It’s not just the code, but also the underlying decisions, workflows, deployments, and signals from production that determine whether an agent can make good choices. To me, that felt like confirmation of something we’ve believed in at Cube for quite some time. For us, GitLab isn’t just a collection of separate tools, but the place where everything comes together.

A few things I found interesting.

• One announcement that immediately got me excited was GitLab Orbit. It brings together context from code, work items, pipelines, and production data into a graph that AI agents can use. As a result, an agent no longer has to sift through an entire repository to understand something, but can retrieve the right context in a targeted manner. That saves time, tokens, and unnecessary work. Orbit is now in public beta, and I’m genuinely curious to see how it develops further.

• I also found the new approach to source code management for AI workloads interesting. Instead of cloning repositories in their entirety, agents can retrieve exactly what they need on the server side. That may sound like a technical detail, but it has a major impact on speed and scalability.

• There was also a lot of discussion about governance. How do you maintain control over AI? Can you see what actions an agent is performing and why? These are questions we grapple with every day ourselves.

• Equally valuable, by the way, were the conversations in between sessions. Researchers from Stanford shared new insights on developer productivity and AI, based on data from hundreds of organizations. And companies like Mercedes-Benz, Google Cloud, and AWS offered a glimpse into how they’re applying these developments. You don’t hear real-world stories like that every day.

The core challenges often turned out to be the same.

What struck me most about this was that I spoke with many different organizations, including companies with thousands of developers. And no matter how big the difference in size, the core challenges often turned out to be the same. How do you measure what AI actually delivers? How do you build trust in what it produces? And how do you ensure broad adoption throughout the organization? These questions are on everyone’s mind, whether you’re a small or large organization.

The difference lies mainly in the approach. Large organizations have to deal with much more compliance and are less agile. You’d often see them set up a separate team focused entirely on this transition, after which broader adoption is managed through that team. At Cube, we’re a lot more agile, so we’re navigating that transition differently. But we’re running into exactly the same challenges. And that’s precisely what made it so valuable to bounce ideas off each other—whether you’re dealing with a company with thousands of developers or a team of forty-five.

Just a quick stop at the Partner Leadership Summit.

Later that afternoon, Job and I also joined the GitLab Partner Leadership Summit. There, the focus was more on the collaboration between GitLab and its partners and how we’re looking toward the future together. Once that was over, we took some time to enjoy London. We went for a walk, had a good meal, and explored the city before heading back to the Netherlands the next morning.

How we handle this ourselves.

What I’ve taken away most from Transcend is that the future isn’t about yet another AI tool. It’s about the foundation on which you run those tools. We’ve been working with GitLab since 2018 and have been an official GitLab Channel Partner since 2024. During that time, GitLab has become the heart of our software development lifecycle.

We’re now building our own AI agents on the Duo Agent Platform—such as the Refinement Agent, which is already in production—and integrating tools like Claude via MCP. GitLab serves as our central hub for context, history, and decision-making. What matters most to me is that AI never operates outside our existing processes. Everything still goes through merge requests, security scans, and automated tests. AI helps us work faster, but in a way that allows us to maintain control.

Looking back.

When I think back on those days in London, it’s not just the stage that sticks with me, but above all the conversations and ideas I took away with me. It was special to be among so many industry experts and to share our vision on behalf of Cube. And to be honest, it was also very energizing to see how much is happening right now in the field of agentic AI. I’m curious to see where we’ll be in a year. If this week made one thing clear, it’s that developments are far from slowing down anytime soon.