Introduction
Welcome to another episode of DJ’s Technology Tuesday. Today it’s the new Anthropic Economic Index. What is that? What’s so interesting about it? The Anthropic Economic Index is a regular study conducted by Anthropic. Anthropic is from the USA, San Francisco. Those who say: “Wait a minute, Jekel, you’ve already said a few times that these are the French, these are the people from Mistral, i.e. from the background there. Sorry again that I’ve already mixed that up here. Yes, Anthropic is the team from Silicon Valley that has a special focus on coding. In other words, they are particularly good and I use them myself to automate programming activities, but not only to automate, but also to prepare for topics. And that’s where we are in the middle of this topic, because the study also looks at the areas of augmentation on the one hand, i.e. the extent to which AI is supported to supplement humans, to augment them, practically, as Miriam Meckel always says, a climbing frame for the brain, or to what extent it is automated. You can see that my motto is a bit like this: “Productivity beyond the hype.”
Global AI usage and country rankings
The euphoria is slowly going down a bit and giving way to what I think is a healthy and good realism. If you look at the whole thing, here, I’ll go to the side a bit. Then you can see in this index, I’ll include the link here, you can see the US usage by specific states. You can see the state usage. The global usage is interesting, because you can see where AI is being used above all else and you can also look at this by country. We are ranked number 28 overall, so if you look at this, you can see the corresponding ranks in which direction it is going. So you can see number one here. Just a moment, I’ll go back to the country usage. You can see that when I look at this, it starts with Israel, Singapore and the United States. You can already see who is in the lead. Israel is also very, very leading in the field of cyber security, as is Singapore. You can also see here that the Swiss are ahead of us, Canada, South Korea, all the nations that tend to be a little further ahead when it comes to digitalization.
European countries, sectors and activity analysis
Yes, we are here, if I look again now, the European countries are slowly coming here. We’re somewhere after Slovenia, but we’re still ahead of Taiwan and ahead of Spain and Italy. And if I take a look at the areas here, you can see what we are doing. Computer Mathematical, Educational Instructional. So you can see a bit of an administrative report. At Anthropic, by the way, it’s not surprising that they have this area, Computer Mathematical, if you look at it. And that’s right, because these models are particularly optimized for that. So that makes total sense. That’s really interesting. I’ll put the link in the comments where you can have a look at what is being done for which area and how. And you can take a look here to see what kind of activities are green in the sense of automated tasks in particular? Where are there tasks that are as augmented as possible? And where are there things that are not included in the data? So that’s quite interesting. Then interesting: You can of course also download the whole thing as a PDF and here you have a little more information on the individual areas.
Notebook-LM: Analysis and visualization tool
It’s definitely worth reading. I have also read it and you can also download it as a PDF. And as I’m sure you’re aware, there’s the option of using Notebook-LM from Google to summarize things like this. I think it’s a great use case, by the way. In other words, I simply put these two sources into Notebook-LM. Publicly available sources, because it’s always called “Data protection and “Google. Hello? This is publicly available information. So, here I just said “Add sources” and then I added these two web links here in the web links that Anthropic put here. What I then did was to say here: “Please give me some infographics via Studio, for one thing. And they’re usually better in portrait format than in landscape format, but I’ve already had things done here in landscape format, once in a somewhat clearer form, once I had an infographic done here in a portrait format in a clearer form. They are usually a bit better and you can then download them as an image, so if you look at these things here. And I think that’s very, very good if you then take a closer look.
Visualization of key messages
You also have the option of making charts here. And I use these charts when I’m giving a presentation, of course, when I’m booked somewhere, then I don’t use these charts. Then they are not at a level where I need them. But it’s quite interesting to simply display the main results from the things here. And let’s just take a look at that. You can also look at it here on full screen or I’ve looked at it here in the window and we’ll take a look at it on full screen. Yes, so quite interesting here is an analysis of global AI usage patterns, productivity effects labor market change shortly before the release of OPUS 4.5. So this from Anthropic and this is one of the corresponding models. Yes, let’s take a look at what comes out of this, and I find it very interesting to say these four key messages here: a return to collaboration, i.e. that a bit beyond the hype, i.e. after this phase of “everything is automated”, the topic of augmentation is becoming more important again. Naturally, if you look at the detailed data, you can see that automation is greater when it comes to the use of interfaces, i.e. APIs.
This is in the nature of things, because they are then used via services such as NETN, for example. And when using the platform, it’s more a question of augmentation. But nevertheless, all in all, it’s moving a little less towards automation and a little more towards augmentation, because we’ve simply noticed that the susceptibility to errors in these processes is simply very high. So also quite interesting, Merkel and math, by the way, I can highly recommend as a podcast, also again in a last episode the discussion to say: Let’s rely far too much on faulty software. And even normal programmed software has bugs and AI software, the question is even bigger. Yes, let’s go back to the global distribution. So it’s the case here that the relevant developing countries tend to use it more for further education, for studying, and in the industrialized nations it tends to be used more professionally. Yes, then we have the issue of the reliability paradox, which I have just mentioned. This means that complex tasks are massively accelerated. However, it decreases with high complexity. And this is due to the high probability of errors and the fact that you then have to test it again.
Upskilling, downskilling and job change
And then here again at the bottom right, the topic of ending the job profiles. So here is the topic, so that there is both upscaling and downskilling. And here’s the example that I’m giving, for example, in the travel agency sector, where planning tasks that are somewhat higher-value are then perhaps even replaced by an AI and then simpler tasks remain in the sense of issuing tickets and things like that. And there are other sectors where it’s more a case of higher-value analyses being carried out by humans because the billow jobs are done by the AI. It’s highly interesting in both directions. In addition to the discussion, which we can certainly take up at some point, there is of course a problem when I put juniors through careers. A friend of ours just visited us at the weekend and he used to be a leading partner at EY and also had responsibility. And he also says: “Man, we used to outsource jobs to India at EY, just like McKinsey has been doing for decades. And now more and more junior work is being outsourced to AI and the challenge is simply that we will run out of juniors at some point.
Complexity and the hype cycle
So this is definitely an issue where we really need to pay attention in terms of the future. Yes, if we move on to the next chart, it’s quite interesting: we don’t just measure users, we measure primitives. In other words, the index analyzes five new dimensions, so-called primitives of conversation. And the point here is to say how complex these tasks are, to say, has the AI – I’ll step aside here – successfully completed the task? To say, how many years of training are needed to understand the prompt? Which use case and how many decisions can the AI make itself? What’s really interesting here is that, if you look at it, it’s moving very, very much more in the direction of augmentation, which I think makes sense to say, okay, there are also some exaggerated expectations here, just as I always have with every technology in the Gartner hype cycle, that at the beginning, of course, it says: Yes, the technology, we can automate everything, we can throw all the people out. And then you realize that you have to invest very, very heavily in error correction. So I thought that was very, very good here, where you say here again the different use cases, so in the more developed industrialized nations more for the topic of work, in the emerging countries more for the topic of education.
But also interesting: they are more hungry for education. So from that point of view, more will happen in the future. Yes, the states in the USA are adapting more and more. The southern states are also following suit. I can say this, I once worked for the company Out of Florida for ten years as managing director in Germany. Here is this gap again. As you can see, I find the quality of the visualization of Notebook-LM quite interesting. Certainly not for a keynote speech that I would give, but just to visualize it quickly for a podcast like this, I think it’s really great, because I just said: “Visualize it for me in the style of my website with my colors.” So that’s why we’re very much involved in this dark blue and black area. Yes, and it’s interesting to say here, the statement, which was also new to me, that if we have tasks here that take over 3.5 hours, then the success rate of these automation interfaces drops to less than 50%. So in the dialog with Claude, the rate remains stable for longer. And why is that? The “human in the loop”, because you simply work with it again and again in the dialog, a bit like a new employee, where I then perhaps also briefly enter into the dialog in between. And that’s better than if I say: “Do the task and then afterwards I’m surprised at what comes out of it.
So that’s really interesting again. Here we have this topic of “upskilling” and “downskilling” again. To say, technical writer was once this topic here, where we also said, okay, automated AI can perhaps automate this topic here, analysis, scope, revision, where you spend 16 to 18 years of education. And then perhaps only sketches and observations, i.e. human activities that are low-skilled, will remain. And on the other hand, here with the property manager, you say: “Okay, accounting, rental contracts, standard things automated by AI, and then you’re left with high skills.” So you’ll see both. So it’s interesting to say what’s left over is what the skill profile is afterwards. And I find it interesting that we said earlier: “Okay, the billo jobs will be automated.” So then this consideration was to say that management tasks, business performance tasks are also possible. And there is an investment company in the United Arab Emirates that has already appointed a board member as an AI or has appointed AI. And the point here is that you say, okay, it goes both ways, i.e. upskilling and downskilling. I found that very interesting in the context of this study.
Reliability as a bottleneck
Yes, so here again this topic, the bottleneck is not the ability, but the reliability. And here also a revision of the estimates, i.e. that one says theoretical potential 1.8%, productivity potential is currently reduced to 1%. And also very interesting, I don’t know if you read the Handelsblatt last Tuesday, the study on the subject of the World Economic Forum in Davos, where PwC also published a study on the subject of productivity. And the result was that 88% of companies there had not achieved any progress in productivity. And in Germany, the figure is a whopping 98%. So only 2% of CEOs, and there were over 4,400 CEOs surveyed in total, said that they had made progress in productivity. So productivity is always beyond “prompt the hype”. So it’s very important to simply take a look at this. And the bottleneck is, as always, the human being and here again what many studies also confirm, to say: “Okay, the smarter the prompt, the better the result.” In other words, if I simply put a simple, uneducated person in front of this box, the result will be different.
This is practically a multiplier of this whole topic. So this is also, naturally, what I said earlier, APIs for structured processes, automation, chat for complex problem solving. The important thing is simply that I say here: people as humans in the loop, especially with over 3.5 hours that you need accordingly and it’s exciting to say that this game has not yet been decided and it’s important that you do two things as a human in the loop. You know it from me: first switch on your brain, then technology and then of course simply use technology.
Conclusion and recommendation
If you are looking for a sparring partner at C-level who is familiar with the depths of the Power Automate system on the one hand and has an eye on these strategic issues on the other, then I would be delighted to hear from you. Until then, I wish you every success, your personal IT coach for managers Thorsten Jekel.
Also available in: Deutsch
