When you democratize technology and you put it in the hands of almost everybody, you empower everyone. AI is going to be the biggest thing in this decade. I would say we are at the very, very beginning of using machine learning really in every way in which we connect with the world.

I've been in the semiconductor industry over 30 years and there were times when nobody knew what a semiconductor was. But today I think everybody at least appreciates that semiconductors and chips can help power numerous things. I think it's just a pregame show. I think we're just getting to the more interesting part where technology is going to have a much bigger role in progressing the world forward.

The potential value of AI is to be able to accelerate our learning cycle. So in some way accelerate R&D. So the willow chip is able to handle a computation in less than 5 minutes that previously on the best supercomputers on the planet today would take 10 septillion years.

So the cloud is essentially the ability for companies of all sizes, all types to run all their infrastructure in a way that feels infinite and like it has no specific place. I know that for a fact that there's no such thing as a franchise value. And so that means every day you have to get up and hopefully you're doing things that are going to be relevant tomorrow.

What we're trying to do is see where the technology takes us, deploy it into the world to actually understand how people are using it, where the risks are, where the benefits are, what people want, how they'd like it to evolve, and then sort of co-evolve the technology with society.

All right, everyone, a warm welcome to us here at NBIM and thank you for taking time out of your lives to spend that with us this afternoon and evening. So we are really happy to have you here. We have a great set of presentations about the technology and security stack that we have.

But first let us introduce ourselves. We will be your hosts this evening. My name is Gunund. I am head of a market exposure delivery team. And what I do is that I have a set of developers together with me that develop our internally built investment platform.

Yes. My name is Shet Len. I'm head of data here at MBM. We manage the data platform and build the data warehouse and provide data for our analysts.

And my name is Selena. I work in the cyber security center as a cyber security engineer. My job is basically just to make sure that the old fund is safe.

Yeah. To kick off this amazing evening, please give a warm welcome to our CTO Vita Brina.

Welcome everyone to us here in the fund. We have really been looking forward to today and planned for a long time. We get a lot of incoming calls and people that want to meet us and want to hear what we are doing in technology and then we found out that why don't we just invite people here and then we can talk about it and we can explore together in the future how we're going to move forward to bring this country also moving forward in this very exciting area.

We see that as an investment firm that lines between technology and investments are becoming more and more blurred, more and more integrated and that gives us a lot of opportunities. So today we are going to talk a bit about what we are doing and sharing with you and then we're going to have a social event afterwards so we can get to know each other. So again warm welcome to all of you.

At MBIM, technology is a key area of focus, not just as a support function for our employees, but as a strategic enabler. We are constantly looking at how technology can improve the experiences for our investment teams.

MBIM has been through over the years a significant technology journey from fully outsourced operations to a modern cloudnative environment. In the next few moments, I'm going to take you through this journey, a journey that made it possible to do all the stuff that you're going to see this evening.

Before 2017, all of our technology was fully outsourced. We had two data centers in Amsterdam. And the reason why we had two was simply because of redundancy in case of one of them went down. But this setup did give us some challenges.

We had we had even a story about a consultant that had to physically go to Amsterdam and connect to the drivers when they were down to get them back up again. So that meant that we had to maintain everything ourselves from replication between the data centers to maintenance and when maintenance and fixed when downtime occurred.

Procurement also had its problems. Something as simple as replacing discs could take months due to delays in the supply chain. So in short, we had to handle every aspect of the infrastructure on our ourselves and this from an operational standpoint was far from ideal.

So in 2017 we launched the ISAC project a project that was significant in our technology journey. It focused on four key pillars. Insourcing we wanted the best talent inhouse and build internal expertise rather than relying on heavily on consultants. security. We focused on strengthening our security posture which was top priority across the entire organization. Automation with fewer in-house developers than consultants. We knew that automation was essential. So infrastructure as code became a core part of increasing efficiency. Lastly, cloud. We went all in on the cloud.

Strong internal support ended up with us choosing AWS as our cloud provider and we started our migration to AWS in 2018 and by 2019 this transformation was complete.

So between 2017 and now MBM has built a modern development environment. An environment that allows our developers to deliver features to our investment teams quickly and efficiently. We have a setup that gives developers the tools, the autonomy, and the support they need to do this fast without compromising quality and security.

Crucially, all of this has only been possible due to strong support from our leadership group. With their push on a cloud first strategy, we now have created foundation for innovation. This has made it possible for us to test new advanced technologies such as AI. We are integrating AI in our daily life to streamline processes to boost our productivity and to make smarter decision. So today you will see all the cool stuff that we have been able to build as a result of this.

So one of the things that we have used this platform for is to build this internal investment platform. So we buy when we can we buy software when we can but we build when we have to when our requirements are so specific and unique that nobody can deliver us the software that we need. So this is also what I spend most of my time on.

But now to present this we will have our UX UI designer Helena Stavo present our internal system in a presentation she has she has called managing complexity through the burden of really big numbers. Helena

hello everyone. So nice to see all of you. I've been really excited to tell you about what I'm working on all day, every day, thinking about when I sleep at night and all of that.

In 1996, two very important things happened. I was born and the first first deposit was made into the account of 1.9 billion Norwegian croners.

As you all know that number is huge today. It is so much more. It is actually Can you read this? Can anybody read this? Hands up. I actually before I had my first interview here, I sat down and studied this number because I was like what if they ask me what this number is? So this is 19,000 billion Norwegian croners. which is so huge that you can't really understand it.

So I've made some small visualizations here to maybe make you understand it. What if we were to cover the entire ground with Norwegian one croners? So we were to sell all of our assets and have 19,000 billion Norwegian croners on the ground. So everywhere you walked you would step on money basically. That would cover this area. So, you could almost work to walk to Bergen from Oslo and walk on money, which is pretty stupid. So, I don't think we're gonna do that.

Or we could stack this money on top of each other and maybe win the space race if we were to climb on top of this money because that would take us 33 million kilometers into space reaching beyond Venus, which is also pretty crazy.

So the number itself is big but the daily operations here are also incredibly enormous in size. So each day we have transactions going through the walls at embim of 450 million croners. That is more than is traded at version every single day and it is a fourth of the state budget that we handle here every single day. Crazy.

What makes this even more complex is that some or most investment firms focus on one portfolio but NBIM as a whole is one portfolio. We want all sub portfolios at NBIM to be doing or performing great. So NBIM is divided into a lot of different portfolios for several reasons. We have different investment and asset classes. we have different sectors we want to be performing well in etc.

So to try to keep track of this it's very important to have an overview of where everybody is investing. This can mean that if I was a portfolio manager for example wanting to buy stocks in Rose Company Co. Then I would check internally first if there was somebody else in NVIM wanting to sell something in Rose Company Co. because this happens a lot. We have different responsibilities here that sometimes you would want to sell something when your colleague would want to buy something. Doing this and having an overview of this actually saves us a lot of money. We're speaking billions here. We usually are speaking billions here but and it makes NBIM the biggest marketplace that you have never heard of.

So all of this we have to manage somewhere and one of the places we are managing this is in Pulis. Pulis is what I work on every day and that is NBIM's main platform for portfolio management and trading.

Before we had Pilates, we had a ton of different tools. What is really, really cool about working here at MBIM is that everyone is incredibly smart. It's kind of daunting to start working here and realizing that everybody knows a lot about everything, which is really cool. So, we have quite a few portfolio managers and traders that code themselves and make supporting applications. And we have in return, we have a lot of different applications that needs to be maintained.

And what is also very particular about MBM is that we are not that many people that you might think. So we have a lot of money but there are not that many people that actually manage this. So what ends up happening is that each person has very, very very specific needs.

So, how do we take all of these different applications and consolidate them into one platform while still keeping the speed and agility for each person that uses it? And speed is the key word here because one click again is worth a lot of money. So, if you spend too much time getting somewhere then that might be a very bad thing.

Yes. So, this is Pilates. Pilates is not just what you see, but it is a lot of work that lies behind the application that I will of course show you afterwards. Pilates consists of a backend that lets developers build their own nodes and endpoints directly into the back end to build front-end applications and or reports directly into one common and collaborative platform.

What is a really key user benefit for this is that we pull all core systems into the same backend which again in return makes it a lot easier for our users to use because they no longer have to go into several different applications to perform everyday actions.

And then on top of that we have the application itself which is built using UX and business knowledge which again is very, very important for scalability reasons. If we think ahead and try to look at the platform as a whole it's a lot easier to have a platform that will live on for many years to come.

We are building everything inhouse and that is very, very, very important. We are not like any other investment firm. It's super important that we can control the entire process and here we actually control absolutely everything. So the close icon in the top left corner could be a smiley face, maybe a sad smiley face. Are you closing me now? Um yes. So we are controlling absolutely everything and that lets us do a lot of cool things. The main feature of Pilates is flexibility.

Each person that uses Pilates we're on firstname basis with everyone really. So I know exactly what each person wants and they are able to get exactly that in Pilates. Previously we had a lot of static tables etc. But now they're able to build their own user interface and pull in components that they can then rearrange as they would like and see exactly what they need when they need it.

All of these components then listen into a data channel where we prefilter the data that they see. This makes it a lot easier for them to actually have a stable and reliable platform or for us to give them a stable and reliable platform because we don't give them everything up front. We have had a rework of the prolitus platform and previously it was very difficult to load the biggest portfolios. It would stop completely and it's not that weird because the main portfolio has over 15,000 rows. So that sort of makes sense that it's difficult to load all of that data. But now we're able to filter that out and provide some speed and agility there.

But it's not only the layout itself that is flexible. We are also making the data itself flexible so that users can build their own reports and mix and match columns and data as they would like. before. Now we've had a very manual process where a developer would have to go in and build a table or a graph or whatever it may be for the for the portfolio managers and traders. Whereas now very soon they will be able to do absolutely everything themselves and pull and drag and drop data as they would like as well.

I'm going to give you a very quick demo of how a normal person or a normal portfolio manager would spend their time in Pilates. This is a sketch because I'm not allowed to show you the real deal from right now unfortunately. So don't base your investments on this because you will be very disappointed.

Okay. So, this is the portfolio table overview. It's a preset that we have made for the portfolio managers. And what most of them use this for is to get a good grip of what are my holdings. So, where are they today? Um where is it opposed to the benchmark because that's what everybody is looking towards. Am I close or far away from the benchmark?

So, right here, this is why I have to stand up here. Now we have the benchmark column it says 100 and then we have the holdings column says 97. So we are minus3 and I want that to get to a zero. So I'm going to write zero in the in the simulations column and then it automatically populates the simulations beneath. And what is really cool about this is that it then also simulates the rest of your portfolio. So portfolio managers are able to build up their own trade programs and see how this changes the portfolio as a whole using these data that is all sent back.

So to summarize, how do we manage complexity? First and foremost, we have to make sure that we are building something that works for everyone while still being able to be a bit more customizable. But most importantly, we have to make sure that it is stable and reliable. If the users cannot trust the platform or trust the data, we have already failed. So having the correct data structure in place and architecture in place is detrimental to have a good working software to be able to handle all of this complexity.

And then lastly is to be able to provide flexibility and scalability. So that would mean in our case giving them the possibility to basically use politis as a building block to create their own perfect idea of what a portfolio and trading tool should be and secondly to have principles in place for them to make it not look bad.

AI has started. AI is the one that I've been thinking about my whole life and wondering, okay, how hard is it? How are we going to get this? Now, it's actually surprised me that we have this acceleration. I think it's going to make everybody richer, but I think it impacts positively impacts poor people the most. It should be particularly true for the democratization of intelligence.

My entire confidence in this generation of AI started when I started seeing from GPD 3 to 35 and that implementation in GitHub. Now, ChatGpt is unquestionably the single best software product the world's ever made. This is the easiest product to use on the planet. anybody's anybody could use it.

All right. So, now we're going to have a quick panel debate about AI and we thought we're going to discuss this from a couple of different angles. So, we're going to discuss this from an organizational level and then bring it down to the individual level afterwards. And hopefully also we can discuss a bit or or compare what MBM has done our efforts and our voyage to what other have done.

And to help us with that we have invited our top technology management and also a couple of external experts to to help us with this. So let's introduce them quickly. So we have uh Alexandre Stansp. He is from grit AI studio and for nearly two decades Alexander has been leading digital transformations but now Alexander is focusing on making AI accessible and practical. So he's building real solutions to real world problems and he does that with passion and determinations which I guess is why you call the company grit.

Uh and then we have also with us Dr. Robert Angels. He is he runs Capgeemini's global AI lab. He has a PhD in AI and he has actually worked with AI since before we had cell phones. So definitely I look forward to hear before the internet even. Yes, definitely. So a long time and before Helena was born.

And then we have Bilit here. Uh Brigittita is a chief technology and operations officer. Uh so she oversees the technology infrastructure that enables us to manage one of the largest portfolios in the world and she has solid experience from service technology leadership from various large financial institutions. But that's not the only job that Brigittita has here. She has another important role. She is the rock and roll voice of NBIM. So when she's not managing our tech and operations, she is the lead singer of our internal house band called Reputational Risk.

Great. And uh last but not least, we have Nikolai. I think five six years ago, you would have to be very interested in art or silos if you had heard about him. But uh but no, I'm not sure if he really needs an introduction as the CEO of NBIM and host of the In Good Company podcast. What you might not know about uh Nikolai is that he's also a serious foodie.

Okay. So in this first part um we're going to look at this more from an organizational level and uh let's start close to home with a question to you Brigittita. So you've been leading our AI adoption efforts uh but from where you sit so overseeing both technology and operations. What has been your approach to introducing a tools and capability to our organization? And also maybe you can touch upon what you think has worked well. Uh what surprised you and if you've kind of encountered any resistance or any challenges.

Um I generally I would just say that uh in MBIM we have this like very powerful grassroot movement when it comes to adopting new technology. So we always have had people that are running uh towards new opportunities. But uh when AI came uh we kind of understood that this is particularly powerful. This is going to really make a big big change. So we decided that we probably need to do it a bit differently this time.

So uh uh we uh introduced more like a comprehensive top-down approach uh when we we started to to look at this and we've done many things. So the first thing we did was to establish an AI team and we really thought that it would be good to have a center excellence uh in the organization and a place that could explore all the new models that are coming in uh and they change all the time. uh we also wanted to make sure they had a good platform uh so that we could oversee all these things in a good way uh and and people come to them and and ask questions and they are the ones that are you know driving uh a lot of the initiatives uh forward.

Um secondly we decide that we need some mandatory courses. So we've made internal uh seven module courses that everybody had to participate on uh and uh we believe that that has been very powerful for the organization and we know that everybody at least is at at some sort of level minimum level.

Um we also uh thought that we need more people in the organization that can help you know drive this as a big force. So we introduced an AI ambassador force of people uh and they received more training than the others. So they are like uh inhouse consultants, smaller consultants that can help out in the various teams to make sure that we have enough people to support.

Uh we set some uh bold statements on on where what we want to achieve. So currently our our um achievement should be uh 20% more efficiency uh and I think everybody knows that in the organization and they link their own performance to this goal. So everybody has to do something.

Um we uh make sure that we talk to uh external vendors a lot like uh open AAI and anthropic. uh we have brought in external expertise like Alexander here he has helped us a lot uh so uh all in all I mean we have tried to just make sure that this is available for for everyone and one of those things that's also super important is that we make tools available easy availability to the people and for us we have called it the three C's which is a claude uh prompting uh engine that anyone can use without any kind of uh programming or um that kind of skills from before. Uh and then we use cursor for those that are more indepth developers that need more more tools uh more extravant tools and then of course uh Microsoft core pilot. So these are threec's and we really push people to use it all the time and we talk about it like all the time and I mean like all the time. I mean we are talking all the time all the time.

Then would you talk about challenges? Um you can think about that a bit. We get them get back to you. So uh so Robert uh through your work at Capgeemini you meet with organizations across all industries across Europe. Uh how does what just described about AI adoption alignment? What do you see?

Well, it's a bit worse than that because we're global. So not only Europe even. Yeah. And I think most of our clients are also on your list of investments. So um but but yeah. No. So I I empathize with what we're talking English. Yeah. Yeah, I empathize with you on top down and knowledge and enablement. So that's very good. But if you look at the the big organizations that we that we serve that are really successful, uh this top down is of course important. You need a top management that is really engaged that is really using it. You also need the top management that is not naive. I mean but because this is risky business.

I mean the one and 2% improvements they are easily to reach with AI. uh but but the 10 20 30% also increases the risk profile that you're running and not all sectors in the world actually tackle that. I mean we have companies making airplanes. Uh I don't want to sit in an airplane that is built with 20% more efficiency based on an AI at the moment. I just don't want to do that. Uh I mean you want to have two wings that are similar.

And um the the thing is that that there is a risk profile on it. And when we look at it, I mean like uh like you said in the introduction, I've been working with AI really from before the internet. I started in 88 with cognitive psychology and artificial intelligence. And I've been always trying to sell this to people like you. And then the answer was like, oh, this is science fiction. Come back next year. And now it is the other way around.

And the the interesting thing is really that um it is a a very cool and interesting technology but we found out that there's two things that are really working well and there's at the moment I mean if you look at all this after chat GPT stuff that comes on the market there's two things that really work well and and they really work well and that is multimodality. I've never seen multimodality working in my life so good as it does now. So, pictures, text, numbers, sound, combine it, throw it together. We've tried it for decades. We never managed. It's It's working well now. I'm impressed. Analysts, reports, graphs, put them together with the text, make some sense out of it. It's good.

Multilinguality. I mean, I've never had a system that could tackle my my language because I speak four languages every day. I have a German wife. I live in Norway. I'm originally Dutch. My children speak Tik Tok English. I mean, you got all these languages blended and and I have a lot of English um Indian colleagues. So, I got all these influences from other cultures and these these models still understand me.

The two things that don't work well that are underperforming no matter what some Alman or Elen are saying is really abstraction and logical reasoning it's not there there is no evidence that's in these models show me the contrary and the other thing is really context and contextual awareness I mean as long as I worked with AI as soon as you train the model and you got models read as soon as you put them in the real world in a real world context where we people are also living and there is some physical things and effects these models are not maintaining world models that let them actually work well enough.

Maybe I should just No, I just have wanted to bring Alexandros into this. So from your perspective working directly with organizations Alexander, so you're focusing on these practical AI solutions just getting things to work. So what's your take on this? Uh what are you seeing as the key factors that uh you know determines whether air adoption actually delivers results and not just like expensive pilot projects? Yeah,

it's a good question. I think uh first and foremost, I think doing something is better than having this wait and see mentality. Unfortunately, a lot of companies still have this sitting on the fence mentality. And I I'm I'm worried that we get derailed from definitions and discussions about uh well, it's not yet performing reasoning to the level that we want. I mean, a year ago, we were comparably at the stone ages when we look at the model performances. So, forget about it. You don't need a PhD in uh neural networks to use AI today. That's the key difference compared to last year. Now anyone can use AI.

So who succeeds? I think it starts with solving real business problems. It's not just saying let's do AI. It needs to be strategic. Doesn't help that the CEO goes to the IT director and say what are we doing about a AI? And the IT director says uh well uh we've turned on co-pilot. Tick the box. No. Solving Monday morning problems. That's what the real value is about, right?

So I think a lot of pilots that fail, they fail because they try to achieve too much. So So honestly, I think it's simple. Try to ship fast, do good enough solutions, get real feedback from users, and then iterate from there. I think that's where you get successful pilots that turn into production.

Yeah. Right. Nikolai, so in your role, you you meet with business leaders from across the world, what what is your uh what do you think are the most important factors for uh adoption of AI?

No, you need to have a a total maniac at the top and you need to I'm I'm serious. This is just the biggest thing that we will ever experience as human beings. It's unbelievable what's going on now because I have there has never been a time in the world where one company who deploy this properly gain a 20% advantage on the other company in one year and another 20 and another 20 and you know when you have like gained 50 to 100% advantage on your competition they will never catch up.

You win they die. You know these companies are out of business. And it can happen on in a threeyear time span now. So you can't wait for anything. You need to be maniacs at the top. You need to totally implement uh across everything you do. You need to have ambassadors just infiltrating your company and you need to have events non-stop and mandatory to get everybody going. I mean this is this is like deadly serious, you know.

I just, you know, I I I I mean, I I hardly sleep these days because I'm just so uh totally super keen that we just uh uh get uh in front of everybody else. And you need to just That's the mentality you need to have. You know, I was in I was in California uh last uh the week before last and you know, I had a had an hour with Bill Gates and he had just spent the whole week talking to everybody who counts, right? Hasabis Oldman, you know, all the guys. And he was just like, it's just like totally mind-boggling. I've been working on this for my whole life. I've never seen anything like it. And the speed is just like the speed is just accelerating, you know.

and stuff we are doing now in the fund. We we couldn't have done it 3 months ago, you know. We are like, you know, we own 9,000 companies. We monitor them real time in terms of headline risks and so on across the world. We monitor them live in 16 different languages and we get notification if we get any headline on any company anywhere. In the past it took us you know days now it's like minutes real time monitoring of the whole world. I mean hello

um you know the way we can deploy capital much quicker the way we can look at uh uh kind of wage packages and proposals. I mean it's just incredible. Why is why not everybody doing this? Well, I hope they are, but you know, in a way also at the same time, I hope they aren't because uh because uh the advantages for the people who really take these on board is just immense.

All right. So, if we shift in our focus from organization to to the individual level and I'm sure many in the audience are wondering about how AI will shape their professional future. So, um coming back to you again and bringing it back home. So as someone who manages them both technology and operations, how is AI changing the skill set and capabilities that you need in your teams?

I think I would start by saying the ability to change and adopt to something new. I think that's the most important thing as Nicholas says and you can imagine how it was to have him back from that California trip when he came back to the office. uh but the ability to change and also uh um understand that you might have worked on something yesterday that you put a lot of effort into and then today might not have that much value. So I think that's one of the most important thing but of course I mean to to have an education with some technology content is beneficial absolutely at least for us in this fund.

So asking for a friend then will you need more developers going forward or less?

I think that we will need developers. Uh of course absolutely I think that we need uh uh developers. We need data scientists. We need uh also in people that are good in investments. I mean it's just like the collaboration between the various skill sets that's important for the future. But again you have to be willing to change. You always have to be willing to embrace what's new and what's coming and that it might or will definitely impact your daily work.

Alexander, you have uh said that AI will amplify our performance or of potential by 10 to maybe even 100 times. How should we prepare uh us here? Uh how should we prepare for being relevant in two three years from now?

Yeah, I think I I I would agree with Big Git there. This is to me very exciting because it's it's about enabling and empowering people that could not develop or solve their own problems before and now anyone can solve their own problems, right? But that doesn't mean that we don't need developers. So I think like the opportunity space is just growing dramatically which means that we could keep the same number of developers or we could double and solve a thousand times more problems at the same time.

But I think it's really naive and I know this room is probably divided. There are those that hate AI assisted coding and think it's just the worst thing that have ever happened and then others have jumped on. But I think it's naive to think that our job as developers will be the same a few years from now. I I think that's that's just stupid to be honest with you. But I think, you know, those that embrace it will have the potential to be 10x or even 100x versions of themselves. And I think we'll see that smaller teams will be able to compete with large teams that are stuck in legacy ways of working. And I think yeah, the future 10x version of a developer are those that embrace AI and become AI prolific, that collaborate with AI and uh yeah, maybe they're a part engineer, a part product thinker, and a part AI whisperer. I think that's the that's the future uh super developers.

Robert, um uh so so why is data engineering so much more important now with AI? Uh we keep hearing that right that data quality and data engineering becomes more

so data engineering has always been important. I mean before CH GPT you built data models on your own data and you put them in production. Now we get global foundation models that are built on basically all data that's available on the world on the worldwide web. So there's no competitive advantage in using that if you don't know how to use these foundational models well and not inject your own data into it. So to your point I mean data engineering will be extremely important. It's your only competitive advantage because everybody can do 1000 and thousand ticks with the same foundational models with the same strategies and know how to prompt.

So you need to add that extra. So I and you see that so India is actually trying to build models they have 22 official languages and they're building models for most of them so that they can enable all their smart people in all these regions in their own languages. Other things are available to everybody. So having your own data available, being able to inject that, build your own models in your own infrastructures with it, that is definitely a thing that we see changing a lot and very fast at the moment and like we've said several times. So at the moment there is new things coming up all the time and we went from chat GPT to rack to graph rack to whatever. Now we're talking about the gentic AI uh whatever that might mean.

All right, Nicolola, I have a bit of a tough question for you also on this. So, bring it on. Yeah. So, this is the elephant in the room. Will people lose their jobs to AI?

Well, for sure the people who don't use AI will lose their jobs. I hope so. I mean, here at MBM, if you don't use uh AI, you'll be you will never be promoted, right? Uh because you will you will lag so much behind everybody else. So, um so will you lose your jobs? Um well um for sure not in the short term I think it's all about uh you know making the efficiency uh better and so on but you know in the longer term there are some uh type of jobs that will be um challenged right uh lawyers accountants uh engineers all these kind of things

and I think that's that's when you look at the time we are at now you'll have uh many type of jobs which will be challenged in the white color space, but at the same time, you know, with the humanoids and so on coming in in the tens of thousands and hundreds of thousands next year and the year beyond, you know, we'll have we have challenges on the white on the blue collar side, too. So, we are um at an incredibly interesting time.

Do you think we'll see like here in MBM, do you think we'll see like fully autonomous digital portfolio managers anytime soon?

No, we are very So, we we never have anything that goes straight from AI to execution or trade, right? there's always a human uh in between and so all the investment decisions here are being taken are being taken by us.

just also a question to the panel about AI maybe adoption. Do you see, you know, generational differences in how people are adopting tools? Is it, you know, just the younger people who are doing this or is it throughout their careers? What's what do you panel think?

Well, I think age has uh obviously plays a big role. I think there have been uh some surveys and statistics about this as well. I think it's maybe more about how they use it, right? So I think uh Gen C more than 50% are using it as a sort of a life partner for any decision and it's almost like they're not making any decisions without first uh uh conferring with chat GPT and I think maybe the uh the millennials and or baby boomers maybe are more in the category of uh impro starting out with improving the speech for their uh their confirmation party or something like that.

But it's really cool what's happened. If you if if you just think about the last 10 years, 10 years ago, who were the heroes in the workplace? The old farts like me, right? The the old people. They were they were really the leading stars. Then five years ago, who were the leading stars? Well, they were the young people, right? Who could code because you needed if you could code Python, you were like a hero. So then all the other people, all the old people felt like losers and the young people, they felt like heroes.

Then Chhatt comes along. Then suddenly it's changing again. No, the old people like me, we can code, right? We don't need we don't need these people anymore. And so it's like it's upside down the whole thing. So basically age has been neutralized which is really cool.

It means another thing being mid mid-level managers. In the old days it meant that you took whatever the guys at the top said and you transmitted it down through the organization. Right now you also function as a root system. You suck up the technology from uh you know from elsewhere in the organization and you transmit it to the top. So being a mid-level manager has totally changed over the last three years. I would say your function has totally changed. It's really fascinating. I mean we are living through some total revolutions now in corporate life. I mean it's so much fun.

Cuz I think we're running out of time here, but we wanted to have like a final rapid, you know, question to all of you. So, if you had to give the audience here one key insight into about AI's impact going forward, what would it be?

Let's start with you, Nicola. Ah, it's changing your life. It's uh upside down. It's just get in there and spend all your time on it. And and if you don't think you got time to really educate yourself on AI, you are just totally not getting it because uh you know, just leave I I come into rooms now leave everything you do. You know, it doesn't matter. It is totally unimportant what you're doing now. just train yourself because you're going to become so much more efficient and so much easier to to learn and train yourself with AI now.

Absolutely. What's your answer? Yeah, this is probably even more interesting because um I don't think you need to understand AI at the moment and you don't need to learn it and if I see my discussions with my clients, I haven't been discussing technology for last four or five months. I think what we're discussing especially with this agentic AI what we're discussing now is really okay I can give you an unlimited amount of employees so what what are you going to do what what are you going to solve what is actually your plan so now it's not even an AI strategy it is what is your strategy because companies need to rethink their whole strategy now and that that is this top- down thinking that needs to come.

How about you, Alexander? What's your I think you have a saying in NBIM AI won't take your job. Someone using AI will. And I think that's spot on the mentality. So if you're early in your career now, that's that's the only advice. Jump on it and learn to be proficient with AI. I tell my kids all the time, use your words. That's your biggest asset. learn how to to communicate and collaborate with AI and then you'll leaprog 10 years uh for those who don't. So that that gap will just increase.

I just have to echo some other people say I just embrace it share use cases inspire the organization and just uh open up because uh it's just fun and exciting and full of opportunities and that's how we have embraced it at least.

Next up is security. And I guess you can imagine that um NBIM is an attractive um target for cyber criminals. Uh so that means that we need to have full control of the security aspect. And luckily we have the best security team in the world. And two of them are going to present for you today. That's Selena and Johanses who going to talk a little bit about how we run our security operations. Please.

What if ma what if the hours used on manual cyber investigations could be reclaimed without sacrificing control and insight? This is a question that we have asked in EMBIB and today Yan and I are going to show you how AI has helped us make this vision a real reality.

In MBIM, we have a cyber response team that run a 247 security and operation center. This means that we are constantly monitoring, analyzing, and responding to threats. What does this mean in practice? Well, you can be in deep deep deep REM sleep when your phone buzzes with a high priority alarm. This is a type of alarm that can that can make it look like the oil fund has just vanished. And let's just say it is expected that we wake up extremely fast.

But here is where it gets interesting because as mentioned in in MBIM we are bringing AI more and more into the loop and in cyber response we aren't just using AI to raise alarms but to contextualize them. And what does that mean? Well, let me take you through an incident.

It is 2:00 a.m. on a Sunday and you wake up from your phone buzzing. A high priority alarm has just come in to our monitoring platform. This is the type of call that spikes your heart rate to 180 before you can even process what's going on, but you immediately know that something bad might be happening.

However, no time to waste. We get up, we get dressed, and we are by our desk within seconds. We log in, we launch our monitoring platform, and there it is. Malware detected on a device. It's vague, but kind of heavy. What malware? What device? Is it a lateral movement? Is this just false alarm? In that moment, everything is a possibility and it's our job to find out fast.

So while our mind goes through all the possibilities and our hands are going through the usual triage steps, something is working in the background. Our integrated AI bot. So this is the alarm in our monitoring platform Sentinel. Malware detected and on your right hand side you see a text block with some info covered over. It is not expected that you read it all but it do hold key information. What has happened? Which system has triggered and in this case it informs about a document containing malicious code.

And this is where our integrated AI bot comes into play. Instead of giving us raw data, it creates a executive level summary of what it believes has happened. This gives us a narrative. It gives gives us a storyline, something that's actionable.

Then it goes deeper. It does a user analysis. Who is the user? Is the user compromised? Is it expected that this type of user work with this type of documents? Then it looks at the document. When was it created? When was it last modified? Who has access it and when? And it takes all of this information and generates a timeline leading up to the alarm. Then it gives investigation guidance. What are the next steps or recommended next steps based on previous cases? And lastly, it gives a tentative conclusion. Does he believe this is a threat or a false alarm?

所以，我们不是从零开始，而是领先几步，有了上下文、方向和对警报更清晰的了解。但我们能相信这些信息吗？简短的回答是：可以，但在人工监督下。重要的是要记住，人工智能只是一个工具，不是最终决策者。

那么，我们的 AI 机器人到底是如何工作的呢？我会把它分解成几个关键组成部分，我认为这些是当今大多数高级 AI 应用的动力来源。首先，也许最重要的一点是所谓的 React 循环（React Loop）。React 循环是一个五阶段过程，大多数能够做事并对事物做出反应的 AI 工作流都遵循这个过程。

这个过程基本上是这样的：你有一个问题，比如“这台机器访问了哪些网站？”或者“这个用户和哪些URL通信了？”这是初始输入。然后，AI 有一系列可用的工具，它可以推理应该使用哪个工具来解决这个任务。接着，它执行其中一个工具，得到反馈，然后触发一个新的提示来分析反馈。然后它又回到思考步骤，再次迭代：“我应该调用一个新工具吗？我现在完成了吗？”它就这样迭代，直到解决任务。

此外，这个 React 循环对于简单的任务效果很好，但当事情变得更复杂时，它就有问题了，尤其像安全事件。你可能开始挖掘，但可能什么也找不到。这是一个有点模糊的问题。这时 AI 往往会陷入一个死胡同，一事无成。因此，除了在 React 循环中运行，我们还有一个所谓的多智能体架构（multi-agent architecture）。

这基本上是我们有一堆不同的智能体，它们有自己的指令集和解决不同问题的秘诀，独立运作。然后我们有一个顶层的管理者智能体，负责生成任务并分配给子智能体。子智能体可能专门分析威胁情报，比如查看恶意域名或IP地址；或者专门查看我们的日志数据、用户或机器等等。你基本上是分而治之，每个子智能体在自己的 React 循环中运行任务，并有设定的资源使用限制。

数据民主化：用自然语言探索数据丛林

我将从一个引述开始。这是一位高级管理成员告诉我的，为了保护隐私，我隐去了他的名字。他说，如果他需要数据，他会向首席投资官请求。首席投资官会把任务交给他的主管，主管会指示一位既懂数据又懂 SQL 的分析师。这位分析师会从我们的分析平台 Snowflake 中提取数据，导出到 Excel，然后通过同样的链条传回去。

这个过程效率不高。如果他忘记了某一列、某个维度或某个筛选条件，整个循环就得重来一遍。这不是因为这位管理者不懂数据或 SQL，而是因为我们在 Snowflake（一个基于云的数据仓库平台）中有海量的数据。

我们从数百个来源（API、数据库、文件、数据供应商等）提取数据。我们使用 Fivetran 将数据导入 Snowflake，在那里我们托管我们的分析平台。我们在那里存储原始数据，包括历史和实时数据。然后我们广泛使用 DBT（Data Build Tool：一个数据转换工具）来连接各个点，将原始数据转换为模型化数据，这些数据再被下游的 Tableau、Python、Streamlit 等多种工具消费。

总的来说，这相当于数百个来源，数千 TB 的数据，数十万个对象。所以 Snowflake 基本上就像一个数据丛林，很难找到方向。然而，我们顶层有一个“狂人”，这很棒。我们还有一个了不起的人工智能团队，他们引入了“3C”工具，这完美地填补了数据与业务之间的鸿沟。

但 Claude 和这些大语言模型（LLM）有一个局限，那就是它们是基于公开可用的、过时的信息训练的。今年冬天，Anthropic 推出了模型上下文协议（Model Context Protocol, MCP）的概念，这实际上只是一个将 LLM 与你想要的任何东西连接起来的框架。

我们已经用 Python 构建了自己的 MCP 服务器，它们可以与 Snowflake、DBT、GitHub 或任何 API 对话。如果你能用 Python 做到，你就能把它做成一个 MCP。这为使用 Claude 或 Cursor 等工具开启了一个新的维度。

现在，那位管理者可以做什么呢？他可以输入类似“我在哪里可以找到关于我们持仓数据的信息？”这样的问题，甚至不需要知道 Snowflake 是什么。然后你可以看到 LLM 开始工作，从所有这些 MCP 和工具中获取数据，收集、抽象，做 LLM 最擅长的事情。

它会提供一些发现或建议。它会发现这是一个使用 Snowflake MCP 的好案例，然后编写一些 SQL 查询，推送到 Snowflake，获取响应，并根据响应创建新的查询，如此迭代直到满意为止。然后，它还会利用这些信息发现这是 DBT 模型，于是从 DBT API 中获取一些质量指标。

最后，它将所有这些来自 Snowflake 和 DBT 的信息连接成一个响应，这是一种检索增强生成（Retrieval Augmented Generation, RAG）技术。它会向你展示一系列模型、表格，包含关键质量指标、负责人、敏感度等。这样，即使不知道 Snowflake 是什么，你也不需要懂 SQL，就能很快找到我们有什么数据以及在哪里。

全员加速：NBIM 的 AI 采纳与培训之道

我们所有人都需要利用人工智能，否则就会落后。人工智能现在是 NBIM 所有部门的焦点。工具已经为每个人准备好了，但如果没人使用，这些工具就毫无价值。

几个月前，Nicolai 给了我们一个任务：将效率提高20%。他不太确定他具体指什么，因为他做的事情和我做的、或者和交易员、投资组合经理、分析师、开发人员做的事情都不一样。所以我们需要用一种有点学术性的方法来处理这个问题。

我们求助于沃顿商学院的 Ethan Molik 教授，他对于我们应如何将人工智能融入工作和日常生活有一些很好的想法。我们尝试采纳他的工作，并将其分解为四个关键领域：工具、项目、流程和习惯。

首先，我们需要工具来让每个人都变得高效。我们选择了 Claude 作为我们的聊天客户端，Cursor 用于开发，Copilot for Office 用于所有与 Office 相关的工作。我们称之为我们的“3C”。

其次是项目。对任何人来说，最好的学习方式就是亲手尝试。我们使用了 Claude 项目。它们是尝试新事物、改进它们并进行概念验证的好地方，而无需任何开发资源。截至目前，我们在 Claude 中有超过27,000个不同的项目。请记住，我们只有700名员工。所以对我们来说，这是个很大的数字。

第三是流程。我们如何工作？作为一个公共机构，我们不幸地相当官僚。我们如何使流程更高效？我们是否有因人工智能的引入而变得不合理的政策？例如，我们有成千上万页的合规文件，这些文件由人们精心创建，但之后却很少有人阅读。

最后是习惯。如果我们要提高效率，就需要改变我们的工作方式。如果我们不改变工作方式，我们就一无所获。但习惯真的很难改变。人们喜欢来上班，做和昨天一样的事情，因为那很舒服。改变习惯对每个人来说都是不舒服的。但那已经行不通了。你需要把人工智能带到你所做的每一件事中，无时无刻。

那么，我们是如何开展这项工作的呢？首先，与人工智能团队、技术部门和我们的人力发展团队密切合作非常重要。在2025年，我们迎来了“技术年”，我们正在加速对人工智能的关注。

毫无疑问，高层定下了非常强烈的基调。这是一种紧迫感，基金里的每个人在绩效管理流程中都会被评估人工智能的使用情况。正如 Nicolai 所说，如果你不使用或采纳人工智能，别人就会超越你。

那么，在“技术年”里，我们具体做了什么？我们启动了一个全面的人工智能大使计划，现在有40位大使。我们举办了“技术日”，这些是充满灵感的日子。我们还为开发人员提供了关于 MCP 和 RAG 的培训。我们刚刚完成了我们的“AI 技能提升”计划，夏天过后，我们还将进行 Vibe 编码的培训。

“AI 技能提升”计划包含七个模块，是强制性的。每个人都必须参加。这是关于我们“3C”工具的培训：Claude、Cursor 和 Copilot。这本质上是关于改变习惯。我们的目标是提高效率，并理解如何开始，以降低改变习惯的障碍。这些课程是小型的，每隔一周30分钟，并且非常注重实践。

那么效果如何？我们成功了吗？我们看到了什么？嗯，在培训或学习与发展方面，你并不总能看到行为或习惯的改变，但在这个项目中，我们实际上可以追踪到。我们看到每个模块之后都有非常清晰的联系。Cursor 或 Claude 的活动都创下了历史新高。我们现在有250名活跃的 Cursor 用户，他们用它来编码，这真的很了不起。到目前为止，我们估计已经节省了1.5亿个工作小时。

展望未来，我们需要保持步伐。人工智能公司之间的竞争异常激烈，他们不断推出新东西，我们与人工智能一起发展。我相信我们几个月前就已经达到了20%的效率目标。Nicolai 和领导层的支持至关重要。但目标在演变。正如 Nicolai 多次说过的，这是一个“石油时刻”，一生一次的机会。人工智能将彻底改变我们的工作方式。

历史告诉我们，真正的效率提升只有在我们停止做某件事或完全不做某项任务时才会出现。洗衣机并没有将洗衣效率提高20%，它解放了时间。洗碗机消除了手洗。计算器终结了手动计算。现在，机器人可以打扫你的房子，洗碗，为你做饭。所以你解放了更多的时间。

对我们来说，同样的原则也适用。我们需要从根本上重新思考工作本身。我们做的很多事情都是“伪工作”（sudo work），这些任务的存在仅仅是因为我们的工作结构，而不是因为它们本身创造了价值。我们填写表格来喂养其他表格。我们参加会议来计划会议。我们创建没人读的报告。效率不是做更多的工作，而是做正确的工作。让人类做人类擅长的事，把其他一切都交给人工智能。