Is Your Data Ready for Agentic AI?Lessons From Doxis Summit 2026

Recorded live at Doxis Summit 2026, Munich

Guests: Roland Gruijthuijsen (TGW Logistics) and Steen Dupont (Natural History Museum, London)

00:00:05

Steen Dupont

Now it's actually becoming a reality where people have to check what the agents are doing, what are they actually performing, what are the activities, and is it worth it in the practical sense?

00:00:14

Will McInnes

Hello everyone, and welcome to a very special episode of The Enterprise Content Show. We're recording live from Munich. You may be listening to this in Jakarta, in Moscow, maybe in Maryland — or maybe you're actually just in the room thinking, what the hell is this guy talking about? We're live from the Doxis Summit. Normally, Franzi and I on The Enterprise Content Show get to hide behind microphones and webcams, but today we wanted to do something different — have an experiment and record an episode right here. So we're living dangerously.

00:00:58

Will McInnes

I like to live dangerously sometimes. This is my kind of danger.

00:01:04

Franziska Thomas

Yeah. No retakes, no edits, no pretending your jokes were funnier than they actually are.

00:01:12

Will McInnes

That's the kind of support I've come to expect from you, Franzi. Thank you. But seriously, thank you all for joining us — and for having a real conversation with these two fantastic guests. Two people living these realities, but in two very different worlds.

00:01:31

Franziska Thomas

Exciting. Please give a warm welcome to our guests: Roland from TGW Logistics in Austria, and the well-known Dr. Steen Dupont from the Natural History Museum in London, who has been a guest on the show before. We're very glad to have you here. Thank you for doing this.

00:01:48

Will McInnes

Yeah, thank you very much, Roland. And thank you, Steen. So, Roland, let's start with you, if that's okay. For anyone who doesn't know TGW Logistics — give us the quick version. Who you are, what TGW does, and what kind of information or process challenges come with the world you're in.

00:02:09

Roland Gruijthuijsen

Yeah. My name is Roland. I'm Head of Enterprise Content Management at TGW. TGW is a highly automated warehouse integration company — we implement logistics solutions for big companies. For example, everyone's aware of Amazon: when you get your package, we're a big part of how that happens. As for Doxis — it's one of the main pillars in our company for content management. All the projects we handle are based on a huge amount of content: drawings, contracts, legal documents, invoices — all of that is handled by Doxis. That's why it's one of our big pillars.

00:03:18

Will McInnes

Very good. Thank you.

00:03:20

Franziska Thomas

Thank you. Same question to you, Steen. Warehouse automation sounds very different from your reality at the Natural History Museum. How would you introduce your world — probably with the same complexity — to our audience?

00:03:34

Steen Dupont

I think the reality is we want to be more like that, actually. So, I'm Steen Dupont from the Natural History Museum in London. What we do is we warehouse life, essentially. We take care of objects and artifacts that describe the universe and life, and how that changes. And one of our main challenges is: we have all these things on their shelves, in their drawers, but people will come around and move them without telling us, or tell us but then move them somewhere they aren't meant to be, or give them new names arbitrarily. Those are some of the challenges we face. The museum itself is a big place with loads of stuff. I myself have gone through a number of roles there — which is why Will sometimes calls me Indiana Jones. I was a researcher, then I moved into building contraptions and inventions, then governance, and now I'm an IT Programme Manager leading the programme that's implementing Doxis — which is hopefully going to make us more like an automated warehouse. That's essentially what we're targeting. So it's great to be here with Roland.

00:04:44

Roland Gruijthuijsen

We could probably offer some tips.

00:04:45

Steen Dupont

Exactly.

00:04:46

Will McInnes

That's what we want to see. We want to go to the Natural History Museum and be automated in our movement around the warehouse. That sounds really good.

00:04:58

Roland Gruijthuijsen

We're going to transport dinosaurs. Sounds exciting.

00:05:02

Franziska Thomas

So we've made the perfect match.

00:05:03

Will McInnes

This is lovely, guys.

00:05:09

Will McInnes

And we promised you a warm and intimate environment — and it's definitely warm.

00:05:12

Franziska Thomas

Getting warmer.

00:05:17

Franziska Thomas

Okay, let's dive in. Roland, I'll ask you first. Was there something that stood out from today, or from yesterday's opening day of the event?

00:05:28

Roland Gruijthuijsen

Yeah. In general: it's AI. What was very exciting — but also a little frightening — was the keynote from Peter yesterday evening. And the message around participating in AI was clear: do it now, not in the future.

00:05:51

Franziska Thomas

Don't wait.

00:05:52

Will McInnes

Quick check, Roland — who here felt slightly frightened after Peter's talk? [Audience response.] Not all of you. Who here felt excited? [Audience response.] Who felt both excited and frightened? [Audience response.] Okay, cool. Back to you.

00:06:13

Roland Gruijthuijsen

I'm not the only one, then. Exactly. The other thing is: AI isn't just a theoretical thing anymore. We have to be careful and figure out which use cases can actually be applied, and look at specific processes to move forward with the technology.

00:06:37

Will McInnes

Love it. And how about you, Steen? Was there a particular takeaway, a theme, a phrase — something that's landed and stopped you in your Indiana Jones tracks?

00:06:49

Steen Dupont

There are quite a few. Some of them are about digital transformation — I think all of us, regardless of what we're doing or what stage we're at, are in a digital transformation. Given everything we've seen and what Peter was talking about yesterday, one thing that really struck me was the infrastructure being built within Doxis around supporting agentic AI. For the past three months — since OpenAI and Claude and similar tools came out in March — we've been playing around with agents and trying to figure out how they work. But I've been doing it at home, without access to any of the company's infrastructure, because we're not letting the beast loose. So seeing guardrails and security put in place within a system that's core to us is massively exciting, because it means we can now start looking at some of the solutions we've been trying to find.

00:07:45

Steen Dupont

We actually have some things we designed in Doxis and went: this is not going to work, it's just too complex. We can now turn around and say: hold on — this can actually be done by an agent, and it'll save more than five minutes. I think agents that do five-minute pieces of work are token-heavy, but agents that save you half an hour to an hour are starting to be genuinely worth the token cost and the effort. So there really is a practical element coming into this.

00:08:23

Will McInnes

And so for you, it's that everyone is in digital transformation.

00:08:26

Steen Dupont

Yeah, I think so.

00:08:27

Will McInnes

That's actually quite a mind-blowing idea.

00:08:35

Franziska Thomas

I'm just thinking about what you said — that it's becoming more practical. I think that connects to the idea of moving from noise to meaning. It's not just hype anymore — it's being applied inside actual business processes. Is that what you're seeing?

00:08:54

Steen Dupont

Yes. A slight clarification on what I was getting at: it's becoming practical in that it's becoming more of a sum game. Before, it was 'let's do this, let's do that, look what we can do now.' It's actually becoming a reality where people have to check what the agents are doing — what are they actually performing, what are the activities, and is it worth it in the practical sense? That's a shift, because it's no longer blue skies. It's reality.

00:09:24

Steen Dupont

People are now taking action and saying: okay, we can do it — but which ones are we going to do? Which ones deliver actual value? And that's a massive shift from just inventing.

00:09:37

Roland Gruijthuijsen

And I think that's one of the main things: you have to figure out what the benefit of the entire process actually is before you introduce AI into it. You can't just take anything and try to drop it into an AI bubble.

00:10:00

Will McInnes

Yeah — you've got to figure out the use case and the value you're trying to create. Who here has had a mind-blowing experience with AI? You don't have to explain it or justify it — but you've done something and thought: that is unbelievable. Quick show of hands. [Audience response.] That's really interesting — fewer than I expected, about less than a third of the room. For me: I had data in our Salesforce CRM and wanted to understand what was hidden in it, what story the data was telling. I asked our internal insights team for help, but they were busy migrating a CRM we'd acquired last year and said they'd be available in three weeks. So I exported a basic Salesforce report as a CSV, put it into our approved OpenAI instance for company data, and had a back-and-forth conversation. In about two hours I had the answers I needed. It's not a glamorous example, but in terms of practical value — to your point, Roland — it was genuinely unbelievable. I would have waited three weeks. I got what I needed in two hours.

00:11:57

Franziska Thomas

Where's the track?

00:11:58

Will McInnes

I don't know where the track is.

00:12:01

Franziska Thomas

Okay. We've been talking about practical, day-to-day work, so let's stay there. When you go back to your regular job next week, what's the very hands-on thing you have in mind? Something you've taken from the summit — a piece of manual work you're wasting time on, somewhere you're not getting the insights you need, somewhere you can't access data? Is there something that pops up?

00:12:31

Roland Gruijthuijsen

Yeah. We were actually on the Doxis Roadshow in Vienna last year, and after that the AI possibilities were shown. When we left, we had a couple of ideas about where we could really use AI in the near future. Based on that, we got in touch with the departments responsible for those processes — and they were genuinely excited to see what we could do. The problem is that until now we haven't been able to move forward because of our infrastructure. But there's a lot of potential, especially in our business, for some powerful integrations.

00:13:30

Will McInnes

That sounds good. So what's the next step?

00:13:34

Roland Gruijthuijsen

Define further use cases, see what else will be possible, then prepare the hardware environment for AI processing. And we need to be careful about governance — what will the company allow and what won't it? We have to get the content in order, think about scalability, and — very importantly — when we in IT identify potential opportunities, we need to have that conversation with the process owners, because they're the ones who will be responsible for it in the future and who have to accept the changes. Change management is, I think, a very, very important part of any transformation like this.

00:14:36

Will McInnes

How do you feel about change management, Steen?

00:14:39

Steen Dupont

I was just about to get there, because I have to contain my excitement when I get back.

00:14:42

Will McInnes

I could see you.

00:14:44

Steen Dupont

So I'm going to come back from this with all these ideas and all these things we can do — but we still need to stabilize our system so that AI can actually get to work. One of the main things we want to do, which I think is slightly different from a lot of other companies, is focus on actually tidying up our data first. Everybody says: good data, good output. Our data needs fixing. So we're going to apply AI the other way — instead of trying to get good outputs from good data, we're going to use AI to fix the data first.

00:15:15

Steen Dupont

One of the things that really popped out is: analyzing where the gaps are, where the similarities are. We have an enormous number of duplicate records. Imagine we have people in our system — everyone has people in their system. But we might have ten Charles Darwins. Everyone knows there's only one Charles Darwin.

00:15:40

Will McInnes

There's only one Charles Darwin.

00:15:45

Steen Dupont

We call this the John Smith problem, not the Charles Darwin problem. But what we can now do is look at all these Charles Darwins, see where they were collected and where they were born, and start figuring out whether they're the same person.

00:15:59

Will McInnes

What if there really are two Charles Darwins?

00:16:00

Steen Dupont

There might be.

00:16:06

Steen Dupont

Hopefully the other one has a different birth date and was collected somewhere different. Then we can say: this 'Charles D' — if that's what they entered — is not Charles Darwin. But it looks like him. So we have to start applying this. And it has a huge impact, because we're analyzing across multiple files and multiple instances in the system. We can ask: this collector — what are they identifying? That used to take an enormous amount of work. We'd have to extract it from the system, run all the analyses, then get it back in — which was a headache. With AI, we can start doing some of those things directly in the system, which becomes incredibly powerful. And then later on — hopefully before next year — we'll also start asking it to actually fix some of the problems. Which is even more mind-blowing for us.

00:17:05

Will McInnes

That's super cool.

00:17:06

Franziska Thomas

But that means you trust the AI — enough to give it the power to correct the data in your system.

00:17:15

Steen Dupont

We trust it enough to perform actions according to how we specify it. And this is where I bring in the guardrails, the application of skills, and the security layer you can apply to agents — which means when you ask it to do something, you know what path it follows and what it's actually doing, instead of just arbitrarily asking it to do something. There's a sysadmin layer that puts those directives in place, and then the users can be more open about how they want to use it. But on the question of trust: do I trust AI?

00:18:00

Steen Dupont

Yes. Because I'm naive, an optimist, and a happy person.

00:18:04

Will McInnes

The optimists die first in the zombie apocalypse. Don't be too optimistic.

00:18:18

Steen Dupont

I will say I've survived quite a few nights alone in a jungle.

00:18:23

Will McInnes

Indiana Jones.

00:18:25

Steen Dupont

And the one thing that actually scared me was a cow. I was in my tent, it was a big —

00:18:29

Will McInnes

Cows are terrifying. More people die from cow attacks than you'd think.

00:18:33

Steen Dupont

I thought it was something else. Not a cow.

00:18:35

Franziska Thomas

You just made that up.

00:18:36

Will McInnes

I did not. I don't have the data. But trust me, cows are dangerous.

00:18:41

Steen Dupont

So yeah — the trust is growing. We can all see it. We are effectively starting to trust AI more and more. And I don't think that's a bad thing, as long as we don't lose track of what we're doing.

00:18:53

Roland Gruijthuijsen

I remember a keynote a couple of years ago at the summit where court decisions were discussed, and the AI decisions were much more precise than the human ones. So from that perspective, I think you can trust it.

00:19:11

Will McInnes

Yeah, it's a great point, Roland. When logic is deterministic, reliable and repeatable, you will get better outcomes than from humans. There's actually research showing that if you're sentenced on a Friday afternoon, you're likely to get a worse sentence than on a Monday morning — because the humans are tired, frustrated, in a hot room they want to leave.

00:19:59

Roland Gruijthuijsen

Your favorite football team has a big impact on court decisions too, apparently.

00:20:05

Will McInnes

Really? That doesn't surprise me. Did they win the night before?

00:20:09

Roland Gruijthuijsen

Yes.

00:20:10

Will McInnes

Good. Then there's a good chance you get a lighter sentence.

00:20:13

Will McInnes

Who here has had an experience with AI where it said: 'You're right to push back — sorry, I got that wrong'? [Audience response.] This is another really important experience. I was researching mountain bikes, got quite a good handle on the subject, and then asked an LLM to compare different models. It started making things up — getting core facts completely wrong. And when you point it out, it says: 'Hey, you're right, thank you for catching that.' No — do not talk to me like that.

00:20:54

Franziska Thomas

Maybe I'll bring us back on track.

00:20:56

Will McInnes

Please.

00:20:57

Franziska Thomas

We've covered a lot — but just to synthesize: what do you think organizations get stuck on the most? You mentioned infrastructure. You mentioned governance. Steen, you said you trust the AI. So what is it? People, process, money, governance?

00:21:16

Steen Dupont

I thought about this. The answer — and it's more than one word — is finding the sweet spot between all of those. I don't think it's any single one of them. Making a big change in just one area is wrong. The really difficult thing is finding where you are and where that sweet spot is for the change to actually work.

00:21:39

Franziska Thomas

Fix the infrastructure piece first. Make sure the foundation is there.

00:21:44

Roland Gruijthuijsen

Get Doxis, then do the rest.

00:21:47

Will McInnes

Okay. We're going to do a short closing round — short answers only, although I know that's risky with this panel. Roland first, then Steen. Complete this sentence: after this summit, I think every organization should pay more attention to...

00:22:19

Steen Dupont

Atomic habits.

00:22:20

Will McInnes

Atomic habits. Say more.

00:22:21

Steen Dupont

You can't do the big transformations in one go. You have to do smaller changes incrementally, because change is inevitable. Good one.

00:22:35

Roland Gruijthuijsen

Building on that — and borrowing from Steve Jobs: think big, start small, and scale carefully.

00:22:45

Will McInnes

That's a really good one. Franzi, your wise words.

00:22:50

Franziska Thomas

When you mentioned that quote, I thought of something I saw in one of the presentations: a collection of information is not knowledge. And then it went further: we have to do the work to turn the information we collect into actual knowledge.

00:23:16

Will McInnes

And the line below that was: a collection of wisdom is not truth. Boom. That wasn't me — that was on the same slide.

00:23:25

Franziska Thomas

Back on track. One final idea that deserves a follow-up conversation from the summit.

00:23:31

Roland Gruijthuijsen

As I mentioned before: do it step by step, and figure out where the biggest benefit is with the lowest risk and cost.

00:23:50

Steen Dupont

Consolidation. We've heard about so much functionality within AI and so many different things that can be done along different pathways — how does it all come together? Because it will, right? We're rushing ahead and doing all these different things. How do they actually combine into one cohesive thing that then branches out? I think there's a very much needed conversation about consolidation.

00:24:16

Will McInnes

Consolidation. Good. And Franzi — what's one topic you think deserves more follow-up?

00:24:26

Franziska Thomas

The AI workflows we can build ourselves in AI Studio. I was super impressed by that — I want to dig deeper.

00:24:35

Will McInnes

Can you ask me what mine is?

00:24:37

Franziska Thomas

Yes — what is yours?

00:24:39

Will McInnes

Governance. I know — it's my least favorite topic. But I think it's where the conversation needs to go.

00:25:03

Will McInnes

Okay. Ladies and gentlemen, we've run out of time. I'm amazed and delighted that you spent this with us — and that Roland and Steen chose to do so as well. If anyone is actually listening on the internet: congratulations, and please do continue with your life. Thank you very much, and goodbye.

00:25:22

Franziska Thomas

Thank you. See you next time.