1 00:00:14,080 --> 00:00:18,200 Hello, this is Eva and you're listening to the In Between Tech 2 00:00:18,200 --> 00:00:21,920 and Trust podcast. This episode gets quite well, 3 00:00:21,920 --> 00:00:26,160 personal and special because it's the first solo episode that 4 00:00:26,160 --> 00:00:29,080 I'm going to record. And I thought it's a good moment 5 00:00:29,080 --> 00:00:32,920 in time because I just come back from a week in Davos. 6 00:00:32,920 --> 00:00:38,160 I've connected with a lot of experts on tech, politics, AI 7 00:00:38,160 --> 00:00:43,640 and also the world in 2026. And so I did want to talk to you 8 00:00:43,640 --> 00:00:48,040 about what's been in conversations and wanted to map 9 00:00:48,040 --> 00:00:51,480 out some thoughts that I had afterwards and also some deep 10 00:00:51,480 --> 00:00:53,560 dives that I thought could be interesting. 11 00:00:56,240 --> 00:00:59,920 This year's World Economic Forum couldn't have come at a better 12 00:00:59,920 --> 00:01:05,319 time because getting into 2026, we've already experienced how 13 00:01:05,319 --> 00:01:08,560 many geopolitical tensions there are. 14 00:01:08,640 --> 00:01:13,800 Also how collaboration is undermined by distrust and how 15 00:01:14,160 --> 00:01:18,680 tech convergence impacts the business sector and how we can 16 00:01:18,680 --> 00:01:22,000 also impact it over the next couple of years. 17 00:01:22,000 --> 00:01:25,600 And I want to come to some topics that were discussed on 18 00:01:25,600 --> 00:01:29,280 the ground because next to, you know, all the political speeches 19 00:01:29,280 --> 00:01:33,040 that you might have heard of, or next to also the arrival of 20 00:01:33,040 --> 00:01:36,960 Donald Trump, of Elon Musk and all other world leaders. 21 00:01:36,960 --> 00:01:41,720 For me, there were four specific topics that I wanted to 22 00:01:41,720 --> 00:01:46,120 highlight to you. And 1st, this was tech, second 23 00:01:46,440 --> 00:01:50,680 were the politics of tech, Thirdly, it was serenity. 24 00:01:50,680 --> 00:01:53,880 And then lastly, it was also the topic of collaboration and 25 00:01:53,880 --> 00:01:55,840 trust. So let's start with the tech 26 00:01:55,840 --> 00:02:00,160 aspects because the technological progress was, you 27 00:02:00,160 --> 00:02:04,840 know, with focus on the impact on society, politics and also 28 00:02:04,840 --> 00:02:08,160 the global collaboration. And it was basically on 29 00:02:08,160 --> 00:02:11,760 everyone's talking point. And some of those you might have 30 00:02:11,760 --> 00:02:13,280 heard. So basically in every 31 00:02:13,280 --> 00:02:18,840 conversation that I attended was talk about agentic systems, AI 32 00:02:18,840 --> 00:02:23,760 models and also how they are adopted for a time to business 33 00:02:23,760 --> 00:02:26,000 impact. It's worthwhile to dig into 34 00:02:26,000 --> 00:02:30,600 those because they will shape of how we also collaborate and 35 00:02:30,600 --> 00:02:32,880 navigate the next months to come. 36 00:02:32,880 --> 00:02:36,320 And I wanted to start with the agentic systems aspect because 37 00:02:36,320 --> 00:02:39,680 they are considered and from the conversations that I had, it's 38 00:02:39,680 --> 00:02:42,480 really about the transformation that they'll bring to the 39 00:02:42,480 --> 00:02:45,320 business world. And there are three profound 40 00:02:45,320 --> 00:02:48,400 reasons of why this transformation is in a way. 41 00:02:48,560 --> 00:02:52,800 So first, they transform the world for knowledge workers and 42 00:02:52,800 --> 00:02:56,280 those who use AI to deliver knowledge based output. 43 00:02:56,360 --> 00:03:00,400 And it means that it can vary from very specific single state 44 00:03:00,400 --> 00:03:04,720 type of output to whole products that can be provided, which is 45 00:03:04,720 --> 00:03:07,400 also one of the second topics that I wanted to talk to you 46 00:03:07,400 --> 00:03:11,440 about because a genetic systems will challenge how we look at 47 00:03:11,440 --> 00:03:13,840 organizations and collaboration in itself. 48 00:03:14,000 --> 00:03:17,480 And you might ask yourself, well, why is that? 49 00:03:17,520 --> 00:03:20,880 And I do think that there are several reasons to it, but the 50 00:03:20,880 --> 00:03:23,920 first one is because agentic systems only work if they're 51 00:03:23,920 --> 00:03:28,000 connected via clear processes, that they need to be designed 52 00:03:28,000 --> 00:03:32,720 towards specific outcomes and they need to be monitored for 53 00:03:32,720 --> 00:03:35,440 performance and also for up to date models. 54 00:03:35,600 --> 00:03:38,960 And while I want to talk to you about the model aspect in a 55 00:03:38,960 --> 00:03:42,480 second, I want to linger a bit with the design part because 56 00:03:42,560 --> 00:03:47,240 designing agentic systems required to be clear here on 57 00:03:47,320 --> 00:03:51,280 what they are designed for. And it can be a specific 58 00:03:51,280 --> 00:03:53,640 outcome. It can be a dedicated way of 59 00:03:53,640 --> 00:03:58,160 working together, or it can also mean that you want to solve for 60 00:03:58,160 --> 00:04:03,000 a specific process, a problem or an outcome that an organization 61 00:04:03,000 --> 00:04:06,120 want to create. Now, if you look into the 62 00:04:06,120 --> 00:04:09,520 heritage of potentially the organization that you might work 63 00:04:09,520 --> 00:04:13,800 for, your processes and the way that you collaborate might have 64 00:04:13,800 --> 00:04:18,160 been designed according to the skills hired according to the 65 00:04:18,160 --> 00:04:22,000 output that you want to deliver. And you're used to collaborate 66 00:04:22,000 --> 00:04:25,640 with the strategy department, with a product department, with 67 00:04:25,640 --> 00:04:29,360 finance teams, HR and so on. And this is a really classical 68 00:04:29,360 --> 00:04:33,720 way of how teams departments are structured and all of them are 69 00:04:33,720 --> 00:04:38,120 feeding into like a specific process, but only like they 70 00:04:38,120 --> 00:04:42,280 cover parts of it with their respective skills. 71 00:04:42,400 --> 00:04:46,560 And now if we translate that into an agentic driven system, 72 00:04:46,760 --> 00:04:51,320 agents will take either very specific parts of a task or, you 73 00:04:51,320 --> 00:04:54,440 know, collaborate closely with yourself, for example. 74 00:04:54,520 --> 00:04:58,840 But they will also be designed as systems overall and will be 75 00:04:58,840 --> 00:05:02,200 built for several agents to form a system. 76 00:05:02,200 --> 00:05:06,040 And you can think of them as a an agentic team that collaborate 77 00:05:06,080 --> 00:05:08,960 and that produce uses and is designed for a dedicated 78 00:05:08,960 --> 00:05:12,440 outcome. So it is required to know what 79 00:05:12,640 --> 00:05:17,240 you're aiming for when designing those agents and also what is 80 00:05:17,240 --> 00:05:19,920 needed to activate it, the conversation on the ground. 81 00:05:19,920 --> 00:05:24,880 And that was really approached this view on how organizations 82 00:05:24,880 --> 00:05:29,000 currently struggle with the design part, because it's super 83 00:05:29,000 --> 00:05:32,240 complex and it also requires quite the clarity of what's 84 00:05:32,240 --> 00:05:36,000 possible with AI models where human expertise need to be 85 00:05:36,000 --> 00:05:40,360 catered to, and also how those systems interact and need to be 86 00:05:40,360 --> 00:05:44,440 organized on an overarching and holistic organizational view. 87 00:05:44,440 --> 00:05:47,280 So the complexity of what I've just described is what AI 88 00:05:47,280 --> 00:05:50,200 leaders and levels have referred to when they talked about the 89 00:05:50,200 --> 00:05:54,280 challenge to adopt AI. And there was this general 90 00:05:54,280 --> 00:05:57,520 consensus that organizations need to be fast on the 91 00:05:57,520 --> 00:06:00,240 implementation. Yet most of them feel 92 00:06:00,240 --> 00:06:04,240 overwhelmed when it comes to the action plan of how that really 93 00:06:04,240 --> 00:06:06,440 is to be done. And most of the conversation 94 00:06:06,440 --> 00:06:09,480 that I've been part of during the day is like focused on the 95 00:06:09,480 --> 00:06:13,000 urgency to prepare for those adoption challenges. 96 00:06:13,160 --> 00:06:16,960 And it's been clear that an action plan is quite individual 97 00:06:17,000 --> 00:06:20,600 to the size and the industry and the delivery model the 98 00:06:20,600 --> 00:06:23,800 respective company is in. Though there has been a clear 99 00:06:23,800 --> 00:06:27,680 call to action to train employees to also be clear on 100 00:06:27,680 --> 00:06:32,040 what skills are required and to activate collaboration on an 101 00:06:32,040 --> 00:06:35,400 organizational level. And interestingly, and this is 102 00:06:35,400 --> 00:06:39,560 also what Satya Nadella, the CEO of Microsoft stated, he said 103 00:06:39,560 --> 00:06:44,000 that AI is a platform shift and not an app nor a tool. 104 00:06:44,080 --> 00:06:47,240 And he described it as the next computing layer. 105 00:06:47,240 --> 00:06:51,600 He said that organizations need to go from auto complete to 106 00:06:51,640 --> 00:06:54,640 agent mode to full project autonomy. 107 00:06:54,840 --> 00:06:58,600 And the point that he made is the winners that adapt the 108 00:06:58,600 --> 00:07:02,480 fastest aren't model builders, but they're the companies and 109 00:07:02,480 --> 00:07:06,960 countries that reorganize work around reasoning, which then 110 00:07:06,960 --> 00:07:10,560 again leads to the fact that an organizational chart becomes a 111 00:07:10,560 --> 00:07:14,520 software problem. So leaders, organizations and 112 00:07:14,520 --> 00:07:19,640 also adapting to AI and tech infrastructure need to look into 113 00:07:19,640 --> 00:07:23,680 the way of how they they approach their collaboration 114 00:07:23,680 --> 00:07:28,520 within the company and prepare their teams and transform them 115 00:07:28,680 --> 00:07:33,760 into, you know, the way that software and platforms can then 116 00:07:33,760 --> 00:07:37,000 be activated. The third part that I wanted to 117 00:07:37,000 --> 00:07:40,160 talk to you about in terms of tech is the use of different 118 00:07:40,160 --> 00:07:43,720 kinds of AI models. And it's been widely discussed 119 00:07:43,720 --> 00:07:47,880 on the ground and also in the past couple of months that an AI 120 00:07:47,880 --> 00:07:51,680 model provides the intelligence to drive overarching output 121 00:07:51,680 --> 00:07:54,640 creation. So currently there are only a 122 00:07:54,640 --> 00:08:00,120 few companies who have the means and the skills to provide that 123 00:08:00,120 --> 00:08:02,280 to the rest of the world with those models. 124 00:08:02,440 --> 00:08:06,080 And this particularly fuels the political and the sovereignty 125 00:08:06,080 --> 00:08:08,200 discussion, which I'll get to you a bit later. 126 00:08:08,200 --> 00:08:11,760 But for now, the design and the transparency of how those models 127 00:08:11,760 --> 00:08:15,920 actually work cost some of the conversations because there are 128 00:08:15,920 --> 00:08:20,400 those who call for world models and those who undermine the 129 00:08:20,400 --> 00:08:23,760 tendency to specified and small models. 130 00:08:23,760 --> 00:08:27,840 So I wanted to talk to you a bit about what are world models and 131 00:08:27,840 --> 00:08:30,160 also what are specified or small model. 132 00:08:30,560 --> 00:08:34,720 And starting with world models, these are like neural networks 133 00:08:34,799 --> 00:08:39,000 that understand the dynamic of the real world and also include 134 00:08:39,080 --> 00:08:43,320 physics and spatial properties. That means that they, you know, 135 00:08:43,360 --> 00:08:46,320 have a lot of data in terms of also the environment that we're 136 00:08:46,320 --> 00:08:51,040 in, and they can use input data to generate videos that simulate 137 00:08:51,160 --> 00:08:55,200 realistic physical environments. The interesting part is that 138 00:08:55,200 --> 00:08:58,440 within a world model, you can then build them for prediction 139 00:08:58,440 --> 00:09:00,640 models. You can use them for style 140 00:09:00,640 --> 00:09:04,640 transfer models or like further reasoning models to begin with. 141 00:09:05,000 --> 00:09:08,120 But from what was discussed in Davos, we're also just at the 142 00:09:08,120 --> 00:09:12,080 beginning of how these application contexts are 143 00:09:12,080 --> 00:09:14,560 evolving. And the benefits are pretty 144 00:09:14,560 --> 00:09:19,240 clear because they extend AI capabilities with a deep 145 00:09:19,240 --> 00:09:24,040 understanding of how spatial relationships are being built 146 00:09:24,400 --> 00:09:28,360 and also how physical behavior in three-dimensional 147 00:09:28,360 --> 00:09:30,560 environments come into existence. 148 00:09:30,760 --> 00:09:34,640 On the contrary, there is the rise of specified and small 149 00:09:34,640 --> 00:09:38,480 models that are available for a particular application. 150 00:09:38,640 --> 00:09:42,120 If I talk about particular applications, you can now think 151 00:09:42,120 --> 00:09:46,640 it's for example, defence tech, it's state related, it's also 152 00:09:46,840 --> 00:09:51,640 potentially government driven. And they are designed for narrow 153 00:09:51,640 --> 00:09:56,960 and predefined tasks that are classified or that also generate 154 00:09:56,960 --> 00:10:00,000 particular outcome. And they then On the contrary to 155 00:10:00,000 --> 00:10:03,240 the world models, they use statistical correlations and 156 00:10:03,240 --> 00:10:07,920 they also are specified models looking into the way of so how 157 00:10:07,920 --> 00:10:12,560 they are focused on particular and high accuracy tasks because 158 00:10:12,560 --> 00:10:16,480 that is where you do not want to be dependent on this one 159 00:10:16,480 --> 00:10:19,920 specific, you know, large scale infrastructure that potentially 160 00:10:19,920 --> 00:10:22,240 doesn't serve in the way that you want it to. 161 00:10:22,360 --> 00:10:26,760 And everyone was very clear that going forward, the 162 00:10:26,760 --> 00:10:29,360 implementation of I will consist of both worlds. 163 00:10:29,560 --> 00:10:34,640 And we do need to decide the context that those models are 164 00:10:34,640 --> 00:10:37,480 applied in and the impact that they should have. 165 00:10:37,480 --> 00:10:41,840 So it's essential to integrate them in the different kinds of 166 00:10:41,840 --> 00:10:44,960 facilities or workflows that we look at. 167 00:10:45,040 --> 00:10:49,360 And that it's also orchestrated how their impact on business, 168 00:10:49,360 --> 00:10:53,040 state, society are being monitored. 169 00:10:53,080 --> 00:10:56,640 And it became clear that leaders and organizations are 170 00:10:56,640 --> 00:11:00,480 responsible to put together the most efficient orchestration for 171 00:11:00,480 --> 00:11:06,720 them and that on a state level, it's required to also dive into 172 00:11:06,960 --> 00:11:09,800 the enablement of exactly this orchestration. 173 00:11:09,880 --> 00:11:13,640 While on a societal level, there are still a lot of trust and 174 00:11:13,640 --> 00:11:18,360 ethical unclarities that haven't been solved and that also now 175 00:11:18,480 --> 00:11:21,000 get the more and more into contradiction of how we will 176 00:11:21,000 --> 00:11:22,960 handle it. But if we linger a bit with the 177 00:11:22,960 --> 00:11:26,800 business sides of it, it became quite clear that the future of 178 00:11:26,840 --> 00:11:30,520 AI model application for businesses is about how they 179 00:11:30,520 --> 00:11:34,920 select and orchestrate those models in their specific 180 00:11:34,920 --> 00:11:37,560 context. And for organizations, it means 181 00:11:37,560 --> 00:11:41,880 that they will have to choose from several different kinds of 182 00:11:42,000 --> 00:11:44,040 specific and different model setups. 183 00:11:44,040 --> 00:11:47,200 And also for organizations, it will impact their operating 184 00:11:47,200 --> 00:11:49,480 model. So they need to stay flexible 185 00:11:49,480 --> 00:11:52,720 and clear on the outcome. They will have to navigate 186 00:11:52,720 --> 00:11:57,160 constant model updates and they will also need to still provide 187 00:11:57,160 --> 00:12:01,080 a stable infrastructure so that it allows some sort of a degree 188 00:12:01,080 --> 00:12:06,000 of flexibility, which then again also needs to be translated in a 189 00:12:06,000 --> 00:12:08,960 whole probably other level of delivery. 190 00:12:09,040 --> 00:12:12,560 But it's also required to think about what do we deliver 191 00:12:12,560 --> 00:12:14,920 towards. And interestingly, and in this 192 00:12:14,920 --> 00:12:19,600 context, it was Alex Karp, the CEO of Palantir, who reflected 193 00:12:19,600 --> 00:12:24,360 upon that as a stress test for organizations and for reality, 194 00:12:24,360 --> 00:12:27,520 because he made the point that it's really about institutions 195 00:12:27,520 --> 00:12:32,400 and that real environments are adversarial, which means that 196 00:12:32,400 --> 00:12:35,520 there is no connectivity, There is still messy data. 197 00:12:35,720 --> 00:12:37,720 We need to talk about permissions. 198 00:12:38,040 --> 00:12:42,720 Where is ethics anchored though AI doesn't fix that, but it 199 00:12:42,720 --> 00:12:47,640 exposes it and essential for organization that we need to 200 00:12:47,640 --> 00:12:52,160 provide an and reliable and auditable layer in terms of, you 201 00:12:52,160 --> 00:12:54,280 know, infrastructure. And when we talk about 202 00:12:54,280 --> 00:12:57,600 infrastructure, this was exactly a recurring topic that came up 203 00:12:57,600 --> 00:13:01,880 as well because everyone was pointing towards the requirement 204 00:13:01,880 --> 00:13:05,640 for it to be interoperable, particularly when it comes to an 205 00:13:05,640 --> 00:13:08,320 organizational level. But if we look in to the 206 00:13:08,320 --> 00:13:11,440 business aspects of interoperability, it means that 207 00:13:11,440 --> 00:13:16,120 tools or AI models can interact with each other and ideally they 208 00:13:16,120 --> 00:13:18,680 can do so seamlessly. Currently, it's one of the 209 00:13:18,680 --> 00:13:21,520 biggest challenges for organization and particularly 210 00:13:21,520 --> 00:13:23,440 also for tech companies. Why? 211 00:13:23,440 --> 00:13:27,320 Because most of the tools work in their respective context or 212 00:13:27,320 --> 00:13:31,440 environment and it's hard for them to be integrated with one 213 00:13:31,440 --> 00:13:35,520 another because there are mechanisms and also how they 214 00:13:35,520 --> 00:13:40,160 deliver an output might not work together, also might not be able 215 00:13:40,160 --> 00:13:45,160 to be activated the way that a person, an organization works 216 00:13:45,160 --> 00:13:47,240 with different kinds of infrastructure. 217 00:13:47,280 --> 00:13:51,360 So currently there are several startups and also in Davos, 218 00:13:51,360 --> 00:13:55,080 there were several ones pitching for an interoperable 219 00:13:55,160 --> 00:14:00,440 infrastructure layer that want to provide exactly this kind of 220 00:14:00,440 --> 00:14:03,800 integration of different kinds of tools, of different kinds of 221 00:14:03,800 --> 00:14:07,560 ambitions, of different kinds of models to also serve as a 222 00:14:07,560 --> 00:14:11,480 general connector and to also provide clarity for 223 00:14:11,480 --> 00:14:16,320 organizations of how they can then orchestrate their AI tools 224 00:14:16,320 --> 00:14:20,400 and models and integrate it into the way of how they produce 225 00:14:20,560 --> 00:14:22,560 output. All of this leads into a larger 226 00:14:22,560 --> 00:14:26,400 conversation that still lingers and that is the increasingly 227 00:14:26,400 --> 00:14:32,480 need for an adoption rate that becomes faster in its time to 228 00:14:32,600 --> 00:14:35,520 business impact. And a lot of organizations still 229 00:14:35,520 --> 00:14:37,880 do not see critical business value generated. 230 00:14:37,880 --> 00:14:42,680 There's still, you know, up to 50% of organizations on the 231 00:14:42,680 --> 00:14:46,280 ground that said that for them, it's super hard to implement AI 232 00:14:46,320 --> 00:14:49,000 and to navigate it in a valuable way. 233 00:14:49,040 --> 00:14:52,080 And I think this is due to two topics. 234 00:14:52,080 --> 00:14:56,400 First, it's like the design aspect that we've already dove 235 00:14:56,400 --> 00:15:00,880 into earlier and it's also the lack of focus on the design 236 00:15:00,960 --> 00:15:06,080 process and on organizational architecture required for AI to 237 00:15:06,080 --> 00:15:08,360 really improve and provide impact. 238 00:15:08,400 --> 00:15:12,840 And the second is like access to skill improvement for employees. 239 00:15:12,960 --> 00:15:15,880 And the latter is one of the critical ones. 240 00:15:15,880 --> 00:15:20,760 And there are two aspects of it. On the one hand side, it's the 241 00:15:20,760 --> 00:15:24,520 provision of time and education material for employees 242 00:15:24,520 --> 00:15:27,840 respectively, for people to adopt and learn to their 243 00:15:27,840 --> 00:15:31,800 required background information. However, it also comes down, on 244 00:15:31,800 --> 00:15:33,880 the other hand, to clear responsibilities. 245 00:15:33,920 --> 00:15:36,880 And one of the main key takeaways that I had during 246 00:15:36,880 --> 00:15:40,800 Davos is that it isn't really clear who provides also the 247 00:15:40,800 --> 00:15:43,880 incentives for people to actually follow through. 248 00:15:44,040 --> 00:15:46,560 And I'm not talking about an organizational level because 249 00:15:46,560 --> 00:15:50,320 every, you know, organization does have and provide a lot of 250 00:15:50,320 --> 00:15:53,960 effort for their employees. It is more about the holistic 251 00:15:53,960 --> 00:15:58,000 view of it because tech companies somewhat point to 252 00:15:58,000 --> 00:16:01,440 government to provide more frames, regulations or also 253 00:16:01,440 --> 00:16:06,240 clear guardrails of how future models will be incorporated and 254 00:16:06,240 --> 00:16:08,880 how then again, employees make use of it. 255 00:16:08,880 --> 00:16:13,160 And because due to the lack of a unified approach, also how to, 256 00:16:13,160 --> 00:16:16,360 you know, distribute the responsibilities when it comes 257 00:16:16,360 --> 00:16:18,920 to implementing AI in organizations. 258 00:16:19,000 --> 00:16:24,280 The transparency aspect is so important, but also the fact of 259 00:16:24,280 --> 00:16:28,000 what actually gets implemented respectively, what output we can 260 00:16:28,000 --> 00:16:30,760 expect. So there's a lot of, you know, 261 00:16:30,920 --> 00:16:35,160 navigation to be done and also distrust to be addressed. 262 00:16:35,360 --> 00:16:39,680 And the these are some of the most critical conversations to 263 00:16:39,680 --> 00:16:41,600 be solved in the foreseeable future. 264 00:16:41,720 --> 00:16:43,320 And who is it going to be solved by? 265 00:16:43,320 --> 00:16:48,600 Well, This is why politics, and it's the second overarching 266 00:16:48,600 --> 00:16:52,600 topic that I wanted to talk to you about, got tied into tech in 267 00:16:52,600 --> 00:16:55,880 so many conversations. Either it was on the back of 268 00:16:55,920 --> 00:17:00,000 geopolitical tensions and due to the large scale resource needs, 269 00:17:00,000 --> 00:17:03,520 but also in the context of whether tech is up or downstream 270 00:17:03,600 --> 00:17:07,400 of politics. And if I think that tech is up 271 00:17:07,400 --> 00:17:10,359 or downstream of politics, you might ask, well, that's this 272 00:17:10,359 --> 00:17:12,280 sounds complicated. What do you actually mean? 273 00:17:12,280 --> 00:17:15,520 Well, at the core, the distinction is about direction. 274 00:17:15,599 --> 00:17:18,920 And more explicitly upstream is when the things are defined, 275 00:17:18,920 --> 00:17:22,079 designed and made possible. So you can think about this as 276 00:17:22,079 --> 00:17:25,240 strategy part, the standardization, infrastructure, 277 00:17:25,240 --> 00:17:27,520 resources, data generation and so on. 278 00:17:27,680 --> 00:17:31,880 And downstream is where things are applied, distributed and 279 00:17:31,880 --> 00:17:34,760 experienced. So you can think about it as 280 00:17:34,920 --> 00:17:38,800 products, services, markets, users and social impact. 281 00:17:38,920 --> 00:17:43,640 And so when it comes to tech, that means that chips, cloud 282 00:17:43,640 --> 00:17:46,640 infrastructure, protocols, training data, open source 283 00:17:46,640 --> 00:17:52,440 libraries and so on, these are upstream, while apps, platforms, 284 00:17:52,640 --> 00:17:57,720 consumer devices, workplace tools are downstream functions. 285 00:17:57,800 --> 00:18:01,000 Now you might have to come to the same conclusion and 286 00:18:01,000 --> 00:18:04,400 technologies both upstream and downstream of politics. 287 00:18:04,400 --> 00:18:08,640 And the direction flip depends on different kinds of layers and 288 00:18:08,640 --> 00:18:11,640 also how we're looking into it. And this is at the core. 289 00:18:11,640 --> 00:18:15,320 And I can tell you it was visible so many times of what 290 00:18:15,320 --> 00:18:19,280 was experienced and why tech leaders were so present at 291 00:18:19,280 --> 00:18:22,960 Devils, which particularly also so was shaped by politicians, 292 00:18:22,960 --> 00:18:24,800 right? It became clear that the 293 00:18:24,800 --> 00:18:28,240 interwovenness of how tech companies currently are both 294 00:18:28,280 --> 00:18:32,160 upstream and define how tech impacts us as well as it is 295 00:18:32,160 --> 00:18:36,200 downstream and shapes the application side of it, is why 296 00:18:36,200 --> 00:18:38,440 we're in, you know, those discussions. 297 00:18:38,520 --> 00:18:41,480 At the center of it all, and ideally on what everyone keeps 298 00:18:41,480 --> 00:18:45,080 asking for is that politics, it's upstream of tech and that 299 00:18:45,080 --> 00:18:48,320 it shapes what kind of technology gets built and by 300 00:18:48,320 --> 00:18:51,800 whom, for example, through regulations, geopolitics, public 301 00:18:51,800 --> 00:18:54,240 funding or standards, laws, etcetera. 302 00:18:54,280 --> 00:18:58,840 So they define incentives, constraints, and the strategic 303 00:18:58,840 --> 00:19:02,120 direction at the same time. And what we're witnessing and 304 00:19:02,120 --> 00:19:05,400 what we've been also discussing in the past couple of months or 305 00:19:05,400 --> 00:19:10,560 years is that tech sits upstream of politics and also at the 306 00:19:10,560 --> 00:19:13,640 political inaugurations. If you just remember last year 307 00:19:13,640 --> 00:19:16,880 when we saw inauguration of President Donald Trump, it 308 00:19:16,880 --> 00:19:20,800 reshapes how politics itself functions through algorithms, 309 00:19:20,880 --> 00:19:24,880 surveillance, decision systems and the finance of it all. 310 00:19:25,040 --> 00:19:29,600 So this is how political behaviour gets shaped and also 311 00:19:29,600 --> 00:19:35,760 how institutional capacities and power asymmetries come to view. 312 00:19:35,840 --> 00:19:39,520 And so of course, there is this feedback loop that is defined by 313 00:19:39,520 --> 00:19:44,280 a recursive relationship. Power roles and power structures 314 00:19:44,280 --> 00:19:48,200 shape upstream tech decisions. And tech reshapes society and 315 00:19:48,200 --> 00:19:50,960 politics itself. And politics again reacts 316 00:19:50,960 --> 00:19:54,240 oftentimes too late to complicated or too far away from 317 00:19:54,240 --> 00:19:59,120 operational reality regulations and redirective elements. 318 00:19:59,160 --> 00:20:04,120 But that then again impacts how tech continues to apply and be 319 00:20:04,120 --> 00:20:05,880 applied. And the discussion of upstream 320 00:20:05,880 --> 00:20:09,120 and downstream tech mostly resonated in devils in the 321 00:20:09,120 --> 00:20:12,440 conversations on energy supply data transparency. 322 00:20:12,520 --> 00:20:15,480 To provision talent enablement and so on. 323 00:20:15,480 --> 00:20:19,560 But interestingly and with the tech firms being so huge 324 00:20:19,560 --> 00:20:23,480 economical conglomerates already, they now also shape the 325 00:20:23,480 --> 00:20:26,840 labor markets, influence political processes and also 326 00:20:26,840 --> 00:20:32,040 concentrate power in a few firms or states, which then again push 327 00:20:32,040 --> 00:20:34,240 tech downstream into political outcomes. 328 00:20:34,680 --> 00:20:37,120 And that again undermines why endeavors, the political 329 00:20:37,120 --> 00:20:41,160 presence and the high attendance of tech CEO's was so omnipresent 330 00:20:41,160 --> 00:20:43,960 this year specifically and interestingly. 331 00:20:43,960 --> 00:20:47,040 And if you look a bit deeper into those who are now first 332 00:20:47,040 --> 00:20:51,120 movers in tech and apply their advantage, they define the 333 00:20:51,120 --> 00:20:54,520 standards and they also shape user behavior and they lock in 334 00:20:54,520 --> 00:20:57,680 dependencies which are hard to reverse them again. 335 00:20:57,680 --> 00:21:01,080 And that is what drives the narratives that are framed as 336 00:21:01,080 --> 00:21:04,920 efficiency driven, providing safety or also how they designed 337 00:21:04,920 --> 00:21:07,800 for user experience. And while these are all, and I 338 00:21:08,120 --> 00:21:11,320 really don't want to downplay it, I do think it's important to 339 00:21:11,320 --> 00:21:14,720 deliver exactly that and to also optimize for growth and 340 00:21:14,720 --> 00:21:17,160 innovation. The conclusion is when tech 341 00:21:17,160 --> 00:21:20,480 shapes political decisions as we've just arrived, there is a 342 00:21:20,480 --> 00:21:24,120 certain amount of transparency needed on how those decisions 343 00:21:24,120 --> 00:21:27,200 are determined. So tech needs to move fast and 344 00:21:27,200 --> 00:21:30,440 at the same time, they also need to be hold accountable of how 345 00:21:30,440 --> 00:21:33,600 they get there. And this then needs to be fed 346 00:21:33,600 --> 00:21:36,960 back into the responsibility conversation we touched upon 347 00:21:36,960 --> 00:21:39,880 earlier and the Nevos. This became essentially the core 348 00:21:40,160 --> 00:21:43,840 but underlying non explicit conversation because tech 349 00:21:43,840 --> 00:21:47,200 companies affect billions of users and like the companies and 350 00:21:47,200 --> 00:21:49,480 revenues and so on. And at the same time their 351 00:21:49,480 --> 00:21:54,160 systems are hard to inspect while only a few firms control 352 00:21:54,160 --> 00:21:56,760 core infrastructure. And that then, shapes the 353 00:21:56,760 --> 00:22:00,160 following. The dependency aspect on how 354 00:22:00,480 --> 00:22:05,040 states rely on private tech for public functions and how they 355 00:22:05,040 --> 00:22:08,800 rely on tech for public functions centers one aspect, 356 00:22:08,800 --> 00:22:12,720 sovereignty. And well, sovereignty for all of 357 00:22:12,720 --> 00:22:18,520 us is quite like the huge word. And in itself, it has been, you 358 00:22:18,520 --> 00:22:22,400 know, a top topic and also in every other conversation for the 359 00:22:22,400 --> 00:22:25,800 past two to three years or so. And on a political aspect, I 360 00:22:25,800 --> 00:22:30,280 mean, this is some, this is where tensions on also state 361 00:22:30,280 --> 00:22:33,320 aspects come into play. But from a tech perspective, 362 00:22:33,320 --> 00:22:36,280 this gets really interesting because in the one hand side and 363 00:22:36,280 --> 00:22:40,480 particularly from a European perspective, the need for 364 00:22:40,480 --> 00:22:46,200 technological progress based out of Europe is like essential to 365 00:22:46,200 --> 00:22:49,720 stay sovereign. On the other side, it was also 366 00:22:49,720 --> 00:22:53,640 Satya Nadella again, who said the topic least talk about where 367 00:22:53,640 --> 00:22:56,760 we most talk about this year. And it's the sovereignty of a 368 00:22:56,760 --> 00:22:59,120 firm. But it inherently became a 369 00:22:59,120 --> 00:23:02,640 statement about competitiveness, value and control. 370 00:23:02,680 --> 00:23:07,360 And you might ask, well, why do we talk about it and why now? 371 00:23:07,360 --> 00:23:11,680 Well, given the interwoveness of tech and politics as we've just 372 00:23:12,040 --> 00:23:16,360 dove into, and also as the economical impact of AI on 373 00:23:16,360 --> 00:23:20,880 society became quite clear and quite the central theme that we 374 00:23:20,880 --> 00:23:24,120 need to also disentangle. And before we do that, we need 375 00:23:24,120 --> 00:23:27,360 to talk about what sovereignty in itself and particularly what 376 00:23:27,360 --> 00:23:32,640 it means in tech and AI. And sovereignty means being and 377 00:23:32,640 --> 00:23:38,000 having the ultimate authority to the side and to make or to say a 378 00:23:38,000 --> 00:23:42,760 final decision, set binding rules and also define the 379 00:23:42,760 --> 00:23:44,280 impact. There are different 380 00:23:44,280 --> 00:23:47,920 perspectives, different actors and also different aims to be 381 00:23:47,920 --> 00:23:52,680 sovereign and it can be territorial, legal, political or 382 00:23:52,680 --> 00:23:54,840 economical. But when it comes to tech and 383 00:23:54,840 --> 00:23:57,640 AI, and in my opinion, there are several layers to it. 384 00:23:57,680 --> 00:24:02,760 The compute energy sovereignty and so the ability to secure 385 00:24:02,760 --> 00:24:07,640 economically viable reliable energy for AI compute over time. 386 00:24:07,760 --> 00:24:11,000 So this is where it gets mostly interwoven with territorial 387 00:24:11,000 --> 00:24:14,720 sovereignty and paired with the mining and raw material aspects 388 00:24:14,720 --> 00:24:17,800 because only a few countries currently have the means to 389 00:24:17,840 --> 00:24:21,640 power and the infrastructure needed for the future AI growth. 390 00:24:21,640 --> 00:24:25,080 And this is also why the current conversation of energy supply 391 00:24:25,080 --> 00:24:27,840 through alternative aspects get so important. 392 00:24:28,000 --> 00:24:32,120 Second, infrastructure and platform sovereignty, which is 393 00:24:32,120 --> 00:24:35,400 one of the most complex ones because it's about dependency 394 00:24:35,400 --> 00:24:38,440 depth and some sub layers to look into it. 395 00:24:39,000 --> 00:24:41,360 We talk about compute sovereignty. 396 00:24:41,800 --> 00:24:44,920 How do you get access to GPUs and cloud contracts? 397 00:24:44,920 --> 00:24:47,960 Then you talk about platform server entities. 398 00:24:48,000 --> 00:24:51,800 Which cloud and ML OPS stack do organizations use? 399 00:24:51,800 --> 00:24:54,640 And then there is also the network sovereignty, which 400 00:24:54,640 --> 00:24:58,480 addresses the latency, routing and resilience of AI to work 401 00:24:58,480 --> 00:25:00,960 with. And these layers particularly 402 00:25:00,960 --> 00:25:03,840 are steered by only a few companies in the world, which 403 00:25:03,840 --> 00:25:07,120 get even more specific when compared to the vast amount of 404 00:25:07,120 --> 00:25:10,360 organizations they provide their infrastructure too. 405 00:25:10,440 --> 00:25:13,520 So this is also the conversation you might have followed along 406 00:25:13,520 --> 00:25:17,400 and most shaped by the term the Magnificent 7 with Alphabet, 407 00:25:17,480 --> 00:25:21,880 Amazon, Apple, Meta, Microsoft, NVIDIA and Tesla and focus of 408 00:25:21,880 --> 00:25:24,840 providing critical infrastructure to it. 409 00:25:25,360 --> 00:25:30,320 And then looking into third, the data server and aspect, which is 410 00:25:30,320 --> 00:25:34,760 the ability to control, protect and strategically also provide 411 00:25:34,760 --> 00:25:39,240 continuous data flows generated by AI deployment and the 412 00:25:39,240 --> 00:25:42,880 securance of ongoing data generations as well as the 413 00:25:42,880 --> 00:25:46,560 control of feedback loops. And looking into the data aspect 414 00:25:46,560 --> 00:25:48,600 a bit more. This is where is this underlying 415 00:25:48,600 --> 00:25:52,320 need for transparency and regulation because it plays into 416 00:25:52,320 --> 00:25:55,280 the way of how they do not have sufficiently the possible 417 00:25:55,360 --> 00:25:59,360 ability to control, protect and provide continuous data flows 418 00:25:59,360 --> 00:26:03,120 for them to be secured in the aspects and respective 419 00:26:03,120 --> 00:26:06,240 environments without being dependent on the tech firm 420 00:26:06,240 --> 00:26:10,200 overall. 4th, we talk about operational sovereignty, which 421 00:26:10,200 --> 00:26:14,880 means that we can audit AI decisions, we can intervene in 422 00:26:14,880 --> 00:26:18,320 real time and we can avoid single point of AI failure, 423 00:26:18,320 --> 00:26:22,200 which is specifically critical and regulated or safety critical 424 00:26:22,200 --> 00:26:24,360 industries. Here again, you could think 425 00:26:24,360 --> 00:26:26,800 about the defense second environment and you can also 426 00:26:26,800 --> 00:26:29,840 look, look into the geopolitical tensions. 427 00:26:29,880 --> 00:26:34,120 But This is why it becomes one of the areas to watch and also 428 00:26:34,680 --> 00:26:38,600 where we need to decide on how to move ahead, particularly from 429 00:26:38,600 --> 00:26:42,080 a European perspective because this serenity currently is not 430 00:26:42,080 --> 00:26:44,280 given. And this is like the main focus 431 00:26:44,280 --> 00:26:47,680 areas of a lot of organizations of where they want to push into 432 00:26:47,840 --> 00:26:54,440 to have this operational focus and ability to steer it in their 433 00:26:54,440 --> 00:26:58,520 respective views and not being dependent on like a third party 434 00:26:58,520 --> 00:27:00,240 vendor. There is the design and 435 00:27:00,240 --> 00:27:03,760 behavioral sovereignty and this implies the the ability to 436 00:27:03,760 --> 00:27:09,240 define what AI optimizes for and to also influence model 437 00:27:09,240 --> 00:27:13,080 behaviour systematically and define the control over the 438 00:27:13,120 --> 00:27:16,640 evaluation matrix. This is specifically the part of 439 00:27:16,640 --> 00:27:20,680 dependency that all companies have on the models that they use 440 00:27:20,760 --> 00:27:24,480 because this is also the value aspect the AI and tech companies 441 00:27:24,480 --> 00:27:28,120 try to secure to ensure their level of intelligence and the 442 00:27:28,120 --> 00:27:32,680 progress their respective large language models to keep and also 443 00:27:32,680 --> 00:27:36,000 to make the race for general artificial intelligence. 444 00:27:36,160 --> 00:27:40,000 The last serenity aspect then again is from a talent and 445 00:27:40,000 --> 00:27:43,280 organizational perspective because organizations required 446 00:27:43,280 --> 00:27:48,360 to the internal capabilities that provide an understanding 447 00:27:48,360 --> 00:27:52,720 and operations and also an involvement with AI systems 448 00:27:53,400 --> 00:27:57,040 without external gatekeepers. If you've now heard the seven 449 00:27:57,040 --> 00:28:00,480 different kinds of types of sovereignty, you might get a 450 00:28:00,480 --> 00:28:03,800 glimpse into why this is so important, but also why it's so 451 00:28:03,800 --> 00:28:09,800 complex to well sort it because we have 7 different kinds of 452 00:28:09,800 --> 00:28:13,800 ways of where we currently are dependent on non European 453 00:28:13,800 --> 00:28:17,000 platforms. We're also looking into the way 454 00:28:17,000 --> 00:28:22,160 of how to enable politicians or to enable also financial flow 455 00:28:22,160 --> 00:28:24,840 across our, you know, or European environment. 456 00:28:24,840 --> 00:28:28,680 When we pair this to the criticality of what we've 457 00:28:28,680 --> 00:28:31,800 earlier looked into and that is the influence of tech companies 458 00:28:31,800 --> 00:28:38,120 on politics, you realize that knowing Texas upstream politics, 459 00:28:38,280 --> 00:28:44,640 knowing that tech also benefits from autocratic tendencies. 460 00:28:44,640 --> 00:28:47,720 This is going to become one of the most crucial conversations 461 00:28:47,720 --> 00:28:50,440 that we're going to have in the next couple of months and years. 462 00:28:50,440 --> 00:28:53,520 It's been core of the discussion in Devils and it's also going to 463 00:28:53,520 --> 00:28:56,640 be a core of the discussions that are going to be move ahead. 464 00:28:56,760 --> 00:29:00,880 And interestingly, on the ground, there have been several 465 00:29:00,880 --> 00:29:04,360 solutions tied to it. On the one hand side, it's been 466 00:29:04,360 --> 00:29:08,080 said to focus on making sure that the auditing and 467 00:29:08,080 --> 00:29:12,040 transformation of organization. And when it comes to the state 468 00:29:12,040 --> 00:29:15,440 of infrastructure, it needs to be very clear of where big tech 469 00:29:15,440 --> 00:29:19,160 companies currently have access to, of how they also are 470 00:29:19,160 --> 00:29:22,480 dependent on the work that the different kinds of organizations 471 00:29:22,480 --> 00:29:26,360 do, which tools and platforms could be good alternatives that 472 00:29:26,360 --> 00:29:28,320 are being provided based out of Europe. 473 00:29:28,480 --> 00:29:31,880 Second, it's all about the development of talent and making 474 00:29:31,880 --> 00:29:35,560 sure that Europeans know and have like a respective skill set 475 00:29:35,560 --> 00:29:38,000 to enable an infrastructure shift. 476 00:29:38,000 --> 00:29:42,760 And this is a point which looks into how education systems are 477 00:29:42,760 --> 00:29:45,680 set up. And then thirdly, it's also 478 00:29:45,680 --> 00:29:49,880 about the unification of a financial landscape and how we 479 00:29:49,880 --> 00:29:53,640 enable investment influx because currently there is a lot of 480 00:29:53,640 --> 00:29:59,000 money that flows into the US and that is also enabling US based 481 00:29:59,000 --> 00:30:01,720 company. Now you might have heard about 482 00:30:01,720 --> 00:30:06,280 the EU Inc proposition that was a lot underline has proposed. 483 00:30:06,280 --> 00:30:10,440 It's like one of the best steps into an implementation part and 484 00:30:10,440 --> 00:30:15,840 also into the way of accumulate more financial means to enable 485 00:30:16,000 --> 00:30:20,400 this infrastructure shift and to also attract talent in the mid 486 00:30:20,400 --> 00:30:24,920 or long term, to sum it up. And I think sovereignty aspect 487 00:30:24,920 --> 00:30:27,120 of like one of the most challenging ones to follow 488 00:30:27,120 --> 00:30:30,760 through. And it also remains to be seen 489 00:30:30,760 --> 00:30:34,480 of how we'll establish as European a mindset again that 490 00:30:34,480 --> 00:30:36,440 provides us with a strong foundation. 491 00:30:36,960 --> 00:30:40,080 And then lastly, 2 topics that I still wanted to talk to you 492 00:30:40,080 --> 00:30:42,480 about this collaboration and trust. 493 00:30:42,600 --> 00:30:45,680 And interestingly, both of them are quite tied to each other. 494 00:30:45,680 --> 00:30:48,680 And that's also why I wanted to connect them with each other. 495 00:30:48,840 --> 00:30:52,480 And when it came to collaboration on a global basis, 496 00:30:52,480 --> 00:30:56,280 but then again, also on an organizational or team basis, I 497 00:30:56,280 --> 00:30:59,560 sometimes got the feeling when I talked to, you know, experts 498 00:30:59,560 --> 00:31:03,200 that there was this unspoken question tying into it. 499 00:31:03,200 --> 00:31:06,360 So what does collaboration even still mean? 500 00:31:06,560 --> 00:31:10,120 So can we trust each other? And we also look into a 501 00:31:10,120 --> 00:31:13,560 commitment that we want to collaborate on a global basis. 502 00:31:13,600 --> 00:31:17,120 And while I know and while I experienced a lot of commitment 503 00:31:17,160 --> 00:31:21,760 to enable global scale collaboration, particularly in 504 00:31:21,760 --> 00:31:26,800 tech, the requirements to cater to individual needs became clear 505 00:31:26,800 --> 00:31:30,680 in many ways. Because if we continue to, you 506 00:31:30,680 --> 00:31:33,800 know, provide this commitment from mutual collaboration and 507 00:31:33,800 --> 00:31:37,880 trust, and also on a diplomatic and business level, we need to 508 00:31:38,000 --> 00:31:41,800 enable it through action. And with that, I wanted to round 509 00:31:41,800 --> 00:31:44,280 up a lot of topics that we touched. 510 00:31:44,280 --> 00:31:48,360 We talked about the impact of tech, the interwovenness of tech 511 00:31:48,360 --> 00:31:53,080 and politics, why sovereignty is so crucial but also so complex 512 00:31:53,080 --> 00:31:57,640 to solve, and how we going to steer ahead with collaboration 513 00:31:57,640 --> 00:32:00,280 and trust. And it's now end of January 514 00:32:00,280 --> 00:32:03,320 2026. What I come home with is to be 515 00:32:03,320 --> 00:32:06,600 very clear that this is a current conversation that needs 516 00:32:06,600 --> 00:32:11,400 to to a context and commitment, being very specific of how we 517 00:32:11,400 --> 00:32:16,120 want to shape the world ahead and making sure that the values 518 00:32:16,120 --> 00:32:19,320 that we have, the ambitions that we want to drive forward are 519 00:32:19,320 --> 00:32:23,440 enabled and not predefined by an infrastructure that we might 520 00:32:23,440 --> 00:32:27,160 use. I am very curious to hear what 521 00:32:27,160 --> 00:32:31,120 that came to your head, but until then, thanks so much and I 522 00:32:31,120 --> 00:32:32,000 hear you next week.