How do you approach cybersecurity in this context of digital logistics and AI?
Albrecht: We have been building cloud-native applications like our control tower for about 15 years, and from the very beginning we put a big focus on cybersecurity. So far, we’ve had no incidents there – and that’s not by accident.
Cybersecurity must be built in from the start. You cannot just put it on top like an umbrella and hope it covers everything. If you do that, there will be holes in your ‘umbrella’ – I can almost guarantee it.
We invest a lot of time and money in R&D to screen third‑party components that we integrate. We follow a very strict process so that we don’t introduce backdoors.
For me, the principle is clear: you need secure by design. It has to start in the R&D department and be part of the whole lifecycle, not something added later.
You mentioned agentic AI and industry-grade reliability. What does “industry-grade” actually mean for these agent systems?
Albrecht: If you move towards agentic systems, you will see agents handling many topics across logistics and manufacturing. The key is to make them reliable and industry‑grade.
In industry, there is no chance we can accept agentic systems that are not on, say, a Six Sigma level of reliability. You need deterministic behavior.
Think of a generic AI example: I ask a chatbot, ‘Give me the five best restaurants in Bangalore.’ It will spit out five names. Are they really the best? Who knows? And if I ask again, I might get a different answer. That’s fine for consumer use, but it cannot work in industry.
In regulated sectors like pharmaceuticals or, increasingly, food and beverage, you need to deliver the best result consistently and be able to withstand audits. That’s the difference between generic AI applications and industry-grade AI applications.
Fast forward, I see industry‑grade agentic applications across logistics and manufacturing, and we definitely want to have a say in shaping that.
There have been high-profile discussions about agents going “rogue” and data exposure risks. How do you think about governance for AI agents?
Albrecht: I actually agree with your idea that we need good human handlers for agents – humans in the loop, over the loop, below the loop.
But it also showed how tricky this can become. If you think about platforms, open models, and the noise around them, it’s clear you need a framework to govern agents – from development through to access and security.
We already spend a lot of money today on data segregation in multi‑tenant environments because you don’t want one user seeing data they shouldn’t see. Now imagine you give an agent free access and it suddenly sees everything. That is absolutely not acceptable.
So you need a systematic governance approach: who can access what, how agents operate, what rights they have. In our case, we frame that under our Intelligence Center X network and analytics governance. That’s our way of securing and safeguarding agentic systems.
What is the key message you’d like decision-makers to take away about AI in logistics and supply chains?
Albrecht: Don’t think in silos. Frame your logistics as part of a value chain. That gives you a much bigger solution space and lets you combine options you would never see if you stayed in a narrow silo.
Use AI with a clear purpose, bring in your domain experts, and design for security and governance from the start. Then AI – especially agentic AI – can help you tackle very down‑to‑earth problems like labor shortages, truck utilization, and end‑to‑end optimization, all in a way that is reliable enough for industry and regulation.
If you do that, AI in logistics becomes less of a buzzword and more of a real competitive advantage in your value chain.
