Advancing Agentic AI: A Conversation with Our Cloud Software Lead

You joined Orbem as a working student. What drew you to the company, and what did those early days look like?

My first connection with Orbem was actually through my thesis. In autumn 2018, I saw Pedro pitch at a startup event at the Technical University of Munich, shortly after Orbem had won the TUM Idea Award. The idea of scanning chicken eggs with MRI and using AI to classify them stayed with me.

When I started looking for a thesis project in 2019, I wanted to work at the intersection of medical imaging and AI, so I reached out to Pedro. Orbem had just moved into a small office in TUM’s incubator building in Garching. There were the three founders and Gilberto (Head of Software), who was already doing his thesis with the company.

What attracted me was the combination of an early-stage startup, fascinating technology, and an ambitious, slightly crazy idea. The welcoming atmosphere was there from the beginning. Much of the company’s broader vision, strategy, and ways of working developed while I was already part of it. I still remember our first stand-ups: we literally stood up in that tiny room and talked to each other. Growing quickly in those first years meant we regularly adapted our teams and processes to fit the company we were becoming. COVID added another challenge along the way.

One of the most interesting parts was the transition from a research project into a product company. We were still proving that the science worked while already building a product around it. Learning to build with that uncertainty was a defining part of the early days.

What does the Cloud Software team actually own, and how does it enable the rest of Orbem?

Cloud Software brings together expertise in software and data engineering with a focus on MLOps. Today, the team works as part of our Machine Learning Platform, which builds the tools and infrastructure that help researchers and data scientists to curate data and develop, train, and improve models more efficiently. Our work spans several internal applications and infrastructure components across the ML application development lifecycle.

Ultimately, we help shorten the path from an idea to a working ML product. When Orbem wants to scan a new type of object or develop a new application, we provide the software foundations that make that work easier and more scalable.

How is your team using AI today? Where has it had the most impact on your workflows?

Because we mostly build internal tools, we can turn an idea into something people can try very quickly. We show colleagues a working prototype, get feedback, and learn early whether we understood their needs. That helps us answer both “Can we build this?” and “Is this useful?”

AI is also changing where engineers spend their time. Implementation takes a smaller share, while defining the problem, choosing an approach, designing the architecture, and validating the result matter more. Engineers increasingly need to think about product vision and act as architects and solution designers.

The challenge is making sure faster development still produces maintainable, scalable software, so we’re building the processes and feedback loops to support that.

For Orbem, this is especially exciting because our goals are ambitious. Speeding up the internal tools and infrastructure behind our products helps us bring improvements to food, fresh produce, and health applications faster and increases our company's impact.

“Agentic AI” means different things to different people. How do you define it, and what does it look like in your team’s day-to-day?

For me, agentic engineering means giving AI agents responsibility for a sequence of tasks toward a defined goal, within clear boundaries and with ways to check their work. In software, that can mean an agent taking a specification, implementing it, and verifying the result. The agent does the work, but people remain accountable for what ships. Engineers set the boundaries and review the results, and they answer for the outcome. Our job is to build the AI Software Factory, the tooling and processes that let agents produce maintainable, scalable code that delivers business value.

Day to day, this shifts our attention toward defining what we want to build, why, and how it fits our architecture. We spend more time aligning with colleagues and users, making requirements explicit, and deciding how we'll validate the outcome.

It also pushes us to formalize what a fast-scaling company usually builds up over time: clear expectations, guardrails, and a shared definition of good. And it keeps us focused on the question that matters most: does this actually create business value?

Projects already move faster, and we can explore ideas that were previously hard to justify. But the quality of the outcome still depends on our judgment and the clarity of our direction. Agentic engineering is transforming our industry, and we see ourselves on a continuous learning journey with it. That's a big part of what makes this work exciting.

Can you walk us through a concrete example where an AI agent took on something meaningful? What did it do, and what was the outcome?

One good example is our current redesign of how data moves from our MRI scanners in the field into the cloud. That data is essential for improving our Machine Learning models.

The existing flow involved several services and components and had grown complex over time. Two senior engineers, working with AI agents, built a much simpler ingestion approach in about a month. It will reduce our operational costs and make the data faster to access and easier to work with.

Those shorter feedback loops matter a lot for Orbem. They help us continuously improve our products while keeping our infrastructure economical as we scale. For me, this is a strong example of experienced engineers using AI to accelerate meaningful work.

‍To close: what advice would you give to engineers who want to grow into technical leadership?

Stay curious, keep learning, and develop your own judgment about what good work looks like. AI makes that judgment even more valuable. You need to understand the concepts well enough to set expectations, evaluate an approach, and recognize when an output falls short. That sense of quality comes from hands-on experience.

Broaden your knowledge too. As AI lets individual engineers work across more areas, technical leaders will need to understand a wider range of systems and decisions. Thinking from first principles helps you handle unfamiliar problems without relying only on established patterns.

For people leaders, firsthand experience with AI matters as well. You need to understand how it changes the work so you can help others learn and adapt. The transition can be exciting but also overwhelming, and going through it yourself helps you support people with empathy and practical understanding. Whether you want to lead technically or manage people, my advice is the same: stay close to the work, take ownership, and keep developing the judgment that helps others move forward.

Join our team

Help build a healthier and more sustainable world.

About the Author

Tobias Vitt, Cloud Software Team Lead.

Tobias leads our Cloud Software team and drives the development of our machine learning platform.