The AI Shift: Can AI help experts get their message across?

金融时报（科技），金融时报科技版块 - 来自金融时报的最新科技新闻与分析。关注

This article is an on-site version of our The AI Shift newsletter. Premium subscribers can sign up to get the newsletter delivered every Thursday. Standard subscribers can upgrade to Premium here, or explore all FT newsletters

Welcome back to The AI Shift, our weekly dive into how AI is reshaping jobs and work. This week we wanted to go deeper than aggregate labour market statistics and forecasts, and speak to a knowledge worker on the front lines of AI disruption. So we sat down with the Spanish economist Jorge Galindo, director of the Esade Centre for Economic Policy think-tank, whose output has been transformed by agentic AI in recent months.

Most recently, Galindo used AI to turn years of painstaking policy research into an animated explainer video on Spain’s housing crisis which has so far been viewed more than 2mn times. His experiences at an AI-forward think-tank provide a fascinating insight into the changes that may be coming to knowledge work in the months ahead.

Below is a lightly edited transcript of our conversation with Galindo about how AI is transforming his work as an economist and communicator, as well as our reflections.

John: Jorge, I’m sure you’re familiar with the framework from Arvind Narayanan and Sayash Kapoor at Princeton, that tasks can be split into three types: ‘decide’, ‘execute’ and ‘deliver’. And so far, ‘execute’ seems to be where most of the AI displacement is happening. It struck me that with your videos the economists are still doing the ‘decide’ part, but then the ‘execute’ part would have been previously handed off to an animator or data visualisation specialist, and now you don’t need to do that. So the same action that makes you more productive is taking away that task, or that economic value, from someone you would maybe have hired or brought in as a freelancer to make the video. Does that feel true to you?

Sarah: Or would you not have done it, in which case this is a new thing that wouldn’t have otherwise existed?

Jorge: It’s precisely that, Sarah. I’ve been consuming Vox.com explainer videos since they started to do this kind of stuff. I always wanted to produce something like that, so when we started [the think-tank] we asked for some quotes from potential providers. It’s not that we couldn’t afford it — it’s that the opportunity cost was too high for us. We had the money but we were not willing to allocate it to that purpose. So this is a video that — in the counterfactual world without AI — basically wouldn’t have existed.

John: The other thing that came through to me from that was someone might think that you just went to Claude and gave it two sentences — here, make me a video about this. But actually you gave it hundreds of lines of your own notes. You pointed it at the book you’ve written on this topic. You pointed it to external data sources. So that feels like this is really AI augmenting or amplifying your substantial existing expertise. AI wasn’t replacing that, it was just allowing you to do much more.

Jorge: I gave it the actual structure of the video. The narrative of the video is mine. It’s not just the book I’ve written on housing, but the fact that I’ve been part of the public debate on housing for years now and it has given me a good sense of what works and what doesn’t in terms of analogies, metaphors, narratives, data points that actually get people engaged. The first graphs it gave me looked like a PowerPoint presentation. I gave it comments like ‘no, I don’t like the graphs’. It gave me more ideas, which I didn’t like at all. Finally, I gave it a mood board, and only after that it gave me something close to what ended up being the final product. Also I’m paying for a [€200 per month] Claude Max subscription. And [the entire video] only used 16 per cent of my weekly limit.

John: So basically that means it cost you something like €8.

Jorge: Yeah, exactly. I mean, the best invested money in my lifetime for having some sort of public voice or influence! Also, to say I’m a heavy user of AI would be a gross understatement. I use it for everything. Beyond my work I use it for planning trips, for help with a home renovation project — for everything. So I think that makes a difference as well. And finally, I have a background in the aesthetic dimensions of communication. I’ve been visualising data for years. I’m obsessed with colour. So that makes me aware of what works and what doesn’t, aesthetically, for the purpose I’m seeking. And finding this mix is not a given.

Sarah: Yeah, and is that maybe a slightly reassuring fact if you’re a creative professional? Because you might see this and say ‘oh my god someone who’s an economist who wrote a book can now produce a beautiful, visually pleasing, highly engaging video, and sure, he wasn’t otherwise going to make a video, but in general, does this mean there’s not going to be demand for my skills anymore?’ But it’s quite unusual to have someone who has a combination of aesthetic taste plus deep subject matter expertise.

John: So what will maybe happen in many companies is the animation is still produced in exactly the same way yours was, Jorge, it’s just that this will be done by an experienced creative. So you’ve still got someone with all the domain knowledge who says “we want this video”, but that is then handed off to a creative who is able to do 10 of these in a week, when they used to maybe do one.

Jorge: Yeah, I think that’s a potential equilibrium coming out of all this.

John: One other question I wanted to ask from the economic perspective is: are you finding new bottlenecks now? Whether in terms of videos or the rest of the work you’re doing at the think-tank. What’s the new bottleneck that is stopping your 10x-ing some aspects of your work becoming a 10x productivity gain overall?

Jorge: It’s the fact that analysis and starting new projects has become so cheap that choice matters more than ever. So for instance right now we have thousands of ideas for videos, and I say, first, I want to choose [which ones to pursue], and second, I don’t want to devalue the video by publishing one video every hour. It’s now possible to systematise this obsession of mine with how excessive regulation has made building more costly and more difficult, which is a typical problem of unstructured, scattered, hard-to-get data that LLMs are very good at gathering and structuring for you. But this is just one of thousands of examples like this, and we’re sometimes like “okay we have to focus”, and that is becoming extremely tough. I cannot spend every single Saturday morning tackling the next iteration of analysis and opening a new project and not finishing another one and losing my Saturday morning that I was planning to spend reading a comic book or just being with my wife — just to mention a few things I would do on Saturday mornings before agentic AI. I had a conversation with an economist here in Spain and he had the hope that generative AI would free up time for us. And I said, it will depend on us. For us types who just love coming up with the next question and answering the next question, I don’t think it’s going to work. Unfortunately, I think it’s going to work the other way. And I’m a bit worried about long-term effects or medium-term effects for teams that are doing that.

Sarah writes

Although the AI discourse remains obsessively focused on the idea of automating away tasks we’re already doing, I found this conversation with Jorge a good reminder that the technology is also enabling people to do new things they simply wouldn’t have done before. What did you take away from it, John?

John responds

I thought this was a brilliant illustration of the shortcomings of the ‘can AI do this task?’ framework. We’ve talked a lot about how Jevons paradox can make automation lead to more output not less employment, and the fact that the quality of AI’s output is still heavily dependent on (and thus multiplicative of) the knowledge and expertise directing it, but it was great to see such clear examples in the real world.

The AI Shift: Can AI help experts get their message across?

Sarah writes

John responds

Recommended reading

Recommended newsletters for you