AI, context, and New Zealand’s data center future

Mark Smith 0:05

The intelligence age is not coming. We're already living in it a maximum, and each week I speak with the people building, leading, and navigating a world being reshaped by artificial intelligence. We explore AI technology, the changing nature of work, and what it takes to build a meaningful career in the intelligence age. This is the intelligence age. Oliver, welcome to the show. Good to have you back on again in my new podcast series. I am keen to get your insights as a state of play from your perspective and what you're seeing in the market across, particularly New Zealand and workforce adoption, uh, the fairs, the data centers that are going on uh or proposed, you know, at the bottom of the South Island. Really interested to get your researched insight in this session. But before we get going, tell us just about the last six months. What's changed for you in your perception, understanding, use of artificial intelligence just since January this year?

Oliver Hartwich 1:06

Well, I think the models have obviously become a lot better, more capable. I mean, that's an evolution really that's been with us for a few years now. But I detected a an acceleration in the last half a year. And I think we're on this really upward slope, uh uh sloping curve now that we've been talking about for a long time. I think when ChatGPT got launched in November 22, I think it was, it had a novelty factor, it was moderately useful, it still made a lot of mistakes, but yeah, you had to really find your workarounds around the deficiencies. And and it improved a bit, and we got new models, of course, but really what we've seen in 2026 so far is completely different. It's a different ball game. It's the evolution towards a more agentic way of using AI, different toolkits, not just different uh large language models, but really the ad advent of things like cowork and Claude and Code, of course, but also work in ChatGPT now. Or if you look at perplexity, perplexity computer. All of these agentic platforms have made a massive difference. And then of course, once you've got really capable models and their agentic versions, you can also bring them together and let them talk to each other and let them correct each other, which is again a really exciting development. So I think actually this year has been peak AI for me so far. But I as I know this is not the end of it, and this is the worst it will ever be. And we'll only gonna get better we'll only get better.

Mark Smith 2:35

I like that peak AI. So true. There's been some interesting uh world events that have gone on. Um, and I'm not gonna talk about the the Middle East, I'm actually talking about uh Anthropic released a model called uh Fable with a off-market back end called Methos, and uh the US government effectively shut the export controls on it down. What I observed around the world, other nation states went, oh my gosh, look how vulnerable we are. We we no longer get to choose whether we've got the best that our money can afford. That's been controlled by the US government. I look at New Zealand, tiny island nation, bottom of the world, and I feel exposed from a what we have choice-wise. If we look across the three hyperscalers operating in New Zealand, AWS, Microsoft, and Google Cloud Platform or Google, none of them have AI compute infrastructure in New Zealand, apart from about three months ago, uh AWS put some infrastructure in, but cannot guarantee that any New Zealand government agency or business compute, data, et cetera, will stay in New Zealand if it uses their platform. Doesn't leave doesn't that leave us vulnerable?

Oliver Hartwich 3:56

Yeah, it does. And I'm I'm not sure about you, but of course when Fable came out, I tried it straight away and I was hooked. And then when it was uh suddenly taking off on a Saturday around lunchtime, I developed the withdrawal symptoms. So it was really terrible. And then of course um came the a period where it wasn't quite clear for how long you could actually use this as part of your plan. I mean they had extended it a few times and um it was a certain degree of marketing going on there, I think, from Anthropic, plus of course the advent of GPT-5.6, which forced them into a different model in the end. Yes, but while nobody knew for how long we would still have it, would it be withdrawn again at the request of the US government? Would it s would uh become unaffordable at some stage because you'd have to buy tokens to use it and it would no longer be part of your subscription? So we cramped a lot into that time where we thought, okay, we still have access to it for at least another couple of weeks. Let's do everything we always wanted to do now. Yes, yes. And uh that was the experience here in the office where all my colleagues basically just uh try to prompt as much as possible just to soak as much milk out of this whole thing as we could. Yes. And then of course um they changed again, and now it's part at least um 50% of your normal plan. And the other thing I think that's changed is actually we've had the arrival of Opus V, which in some ways is comparable to Fable. So actually there's not even that much need anymore to go for Fable. It's been a wild ride. And yes, back to your question, it's demonstrated actually how dependent we are on these tools. Or actually, are we dependent, or did we just get hooked? I mean, we used to do a lot of the things without only about four years ago, but we've just realized just how powerful all of this stuff really is. By the way, the other thing I would say is at the beginning of the year I had a brief period where I was actually more into Gemini. Because at the time they were, I think, narrowly in the lead on some of the issues, and I got better answers out of Gemini. I remember actually people telling me, Oh, I'm still in ChatGPT, and I kind of um felt pity for them. I thought, well, you you poor bugger, you're still there and you've completely missed the boat on Gemini and Claude. And last few weeks actually the pendulum has swung back so much to GPT with work and 5-6 Sol Ultra, my goodness. So it is also interesting just to see it from a competitive competition point of view. There is never a moment really where one of the competitors is completely outrunning all the others. It usually takes a week or two and then you've got a level level playing field again. It's really exciting.

Mark Smith 6:30

How do you think about context when you're using multiple providers? In other words, you do a whole bunch of work, let's say, uh with Opus or Fable, and then you're like, at the moment, Sol's performing better over on OpenAI side of things. But all your data and all the work and all the kind of context it understands with you is over on Othropic. And then you get the same whiplash type effect on when you when when it switches. And we know this is always just gonna be a uh an incremental one-upmanship race as we continue on this journey. How do you reconcile that across the platform's context? In other words, not having to re-explain yourself to a new AI when you've already explained yourself or has context of your working history in the other model.

Oliver Hartwich 7:24

Very good question. I mean, sometimes you want the AI to know everything about you and your context. Sometimes you don't actually. Sometimes you want a pair of fresh eyes, sometimes you want another AI to just look at the work you just produced and tell you what it makes of it without actually taking into account everything that you told it before. Yes, so you balance this, I think. In my case, what I've done, for example, I have actually got my own website scraped through Claude Code. I condensed everything into a Claude Skill that knows everything about me, has all my writings, knows how I write, and has my entire back catalogue in in the background. So I condensed all my reports or my books, all my articles, all my newsletter articles, everything into markdown files. Really nicely compressed. It's in it's actually quite interesting to see how much of a life's output you can compress into 14 megabytes if you compress it properly. And there is this clause skill that I have now that basically tells me what I would think about issues, which is quite useful. I've used it for a current project where I just want to make sure that I'm consistent and that I'm not contradicting myself. And so it went through my entire back catalogue to find out things that I said in the past, which I'd almost forgotten. It was interesting. Actually, fortunately, I was largely consistent with myself, but still it is interesting. So uh sometimes what I do is actually I develop filters also or skills in Claude and then transfer them over and into ChatGPT and say, Well, okay, I did this with Claude. I would like to have something similar here, but actually how could this be improved? And I've had this recent experience where I played ping pong between ChatGPT and Claude and developing a skill. Yes, so now I've got exactly the same skill in both, but each round of the ping pong made it better and more comprehensive.

Mark Smith 9:06

Yeah. That's I I I find peer reviewing, one AI peer reviewing the other, is is brilliant for finding gaps, nuance, uh, different perspectives. And you know, I had an issue yesterday with my Clawed desktop, and uh I'm on the max plans of all of them. In fact, I've got nine different API connections in from different providers into what I do, and I used Codex to fix Clawed Desktop from accessing various service endpoints, etc. And the reason is Codex is really good at computer control, it can jump into the settings right down on my computer and go, okay, it's a config file, it's wrong. I don't have to poke my way through a UI. It goes in and goes, huh, here's a gap. There was a tilder or a character on the end of a line, a string or whatever. Boom, I've sold. Oh, it's taken me ages to find that. Yep, it's a good idea. I use it in the same kind of way. Yeah, yeah. Yeah, exactly.

Oliver Hartwich 10:02

By the way, and another really interesting thing that happened in the last half a year, I think MCP protocols are becoming more commonplace. Yeah. And I use one in particular, and that's uh to a tool called Site S-C-I-T-E.ai. I'm not sure whether you've come across them. No. So it's a library giving you access to, last time I checked, about 170 million academic papers in full text for a modest subscription of I believe it was $179 a year, something like that. Wow, that's cheap. And then you can integrate it via an MCP server into Cloud and Perplexity and Chat GPT, whatever you like. And now you can actually peer review effectively your own writing and say, okay, against the kind of bulk of academic literature out there, just tell me, does my text actually stack up? What contradicts, what supports goodness. I mean, if you're an academic, you're trying to do that yourself, going into a library, trying to find the right kind of journal articles on the topic you're writing about, you will spend years doing that. And here you can do it via an MCP server, you link it into your AI and you have it done automatically. This is amazing.

Mark Smith 11:14

I was speaking to uh somebody at CT in New Zealand that's involved for an AI strategy as a uh across the board. And interesting to hear that a PhD in the first year, you're gonna spend about your first year just collating everything that's already been written or you know, on the topic, etc. Now I'm speaking from someone that's not gone through the university system at all. And he was saying that that would be a year of legwork that you would do to bring all that body of knowledge together. And he said, now it's under four weeks and you can bring that body of knowledge together. Yes, you still got to go through it, etc., and understand it. That but the actual he goes, that's one year of a PhD program condensed down to one month of normal legwork that you'd be doing.

Oliver Hartwich 12:05

Yeah, that's right. I did my PhD in law. When was that? From 2000 to 2004. So four years of my life. And yes, the first year looked very much like that. I I was writing a bit too. Yes, but I spent a lot of time in the library. I photocopied loads of journal articles, I had entire boxes full of stuff at home that I could then help myself with. Well, these days you would condense that. You would have access digitally already. I mean the only disadvantage in all of this is of course that extra year that it takes you to collate stuff also gives you an extra year to think about it. I think, yeah, exactly. And it's not just um the time that you spend um in the photocopy room at the library, it's actually what you do with the stuff afterwards. You take it home, you read it at night, you get a night's sleep, you think about it again in the morning. It gives you time to actually form the thoughts in your brain. I'm not sure whether that would still work in the same kind of way if all the stuff is automatically there, and then you give yourself four weeks to actually get on top of all of this. I don't think it would work that way. Yeah. That's actually where all the AI tools just reach a kind of a capacity threshold within the human brain. We probably have to find a way of dealing with that.

Mark Smith 13:14

Yeah, it it it is, and you're so right that you know, if I've got a hard problem, I find sleeping on it is a great way to to come up with a solution. In fact, I I remember a guy working at my property in the early days and he was doing some earthworks and run into a a problem with a power cable in the ground. And he goes, You know what? I'm not gonna do anything about it today because the accidents just happened, wasn't exposed or anything, as I well, it was physically exposed, but not a a hazard. And and within a couple of days we came up with a solution to solve it, but it you could either freak out right then and try and solve it, or what happened is that sleeping on it, thinking about a few things, running some ideas past some other people, come up with a solution. I think that percolation time the brain needs to often tap into, you know, uh one-on-one make five type thing, you know, the the the the pattern.

Oliver Hartwich 14:12

And I believe there's good research and evidence on that. The brain actually really solves problems during your sleep. So actually that's something to keep in mind while we're all busy in front of our screens using AI tools all day and all night. Every now and then you probably just need to give yourself a bit of a rest to actually have a chance to process all of the stuff.

Mark Smith 14:30

Yeah. Have you been doing any research in or or or around the AI data center discussion in New Zealand?

Oliver Hartwich 14:40

A little bit. Well, let's put it this way. I haven't actually done proper research, but I followed it, of course. Yes. The the thing about the data centers in New Zealand that always puzzles me is actually that I think we could be the data center superpower in the world if we wanted to. We have endless energy under our feet. I mean, being in New Zealand comes with a few disadvantages, of course. It's the shaky isles. Every now and then we get natural disasters, volcanoes, earthquakes, but the same source of energy that causes all of these natural disasters is, of course, a source of energy we could also tap in the good times to power our data centers and indeed everything else. So I think the only thing standing between us and becoming a data center superpower is actually our ability to really fast track all of these applications and really build the geothermal installations that we need. Often that takes too long. Often there are regulatory hurdles, but in principle there's nothing that should stop us. Except I would say on a morning like today, when it was bitterly cold in Wellington. Not sure how cold it is where you are.

Mark Smith 15:46

It was cold, cold, cold as it's ever been in six years, I think today for me.

Oliver Hartwich 15:50

Yeah, yeah, no, it felt like that here too. Um and of course we have a warning from Transpor that they might run into some issues. And yesterday actually we got another warning that um by the end of the week, I think they said at 8.30 in the morning, New Zealand might exp uh expect some kind of crunch time in the um transmission network. It is hard when it when we have this situation to then make the case, hey, let's um just build a few more data centers, and of course it will be extremely power hungry. We have to, in a way, make sure that these data centers that we would like to build have social license, and that only works if we make sure that our network as such is more stable and more reliable.

Mark Smith 16:32

So so yesterday my power went out at 9 a.m. I had uh and didn't come back until three from the North Power provider. I happened to have put solar in November last year, and I've got full battery backup, and so didn't lose a glitch, just you know, steady as you go right through by putting that redundancy in place for my own business. Isn't there an opportunity to really lean in? I know there's been a lot about solar in the elections, and I'm saying rather than the state or the the government-sponsored utility companies uh providing us a service, why don't they just allow us to provide our own utility, make it really attractive to provide our own utility, and then with battery storage placed all across their network, I'm saying I'm talking about container load size batteries, take that high efficiency load that comes midday from the sun, bank it in the batteries, and then when we hit peak time in the evening, in other words, level out the the the curve, right, on consumption, and so therefore bank it back into the grid. And so yeah, but it takes the actual network infrastructure to think differently about a future power network than the current one.

Oliver Hartwich 17:47

Yep, I think that's already happening. I mean, we need a lot more of that, of course, but we have a situation where solar has become much more cost competitive. Where I think it doesn't really require too much of a subsidy anymore to be competitive. And so we can expect a much bigger rollout of solar in the coming years, and yes, combined with batteries, that will solve problems like yours. So you wouldn't actually be dependent on the grid anymore, and you could actually take care of your own needs. The only thing I'm a little bit skeptical about is actually will that also help us with the seasonal fluctuations? I mean, solar power and battery power then are very good ways of actually smoothing your energy consumption out over the day for yourself. Yes. But we know that we've got seasonal patterns in New Zealand as well. And I'm not sure whether the installed battery capacity would really make a meaningful dent into that.

Mark Smith 18:42

It comes down to it doesn't have to be a hundred percent one thing or the other, right? But you talk about thermal, you could combine these things to actually create a robust or resilient infrastructure.

Water Consumption in Data Centers

Mark Smith 18:52

So one of the things we've heard a lot about is in the data center space is the use of water or water consumption. And I know based on the South Island deployment, that's been a big thing about their need to harvest. And I think that sometimes you don't do a comparison of a number to another number or another industry. And so, you know, some of the data I've seen that a a modern closed loop uh cooled system for a modern data center uses the equivalent of maybe two McDonald's restaurants in their water consumption a year, or a fifth of that used by a golf course. You know, an average-sized farm in New Zealand would probably be dairy farm, would be milking two, three, four hundred cows. A cow drinks about 70 liters of water a day and is consumes about 70 liters of water per washdown after, you know, the the milking time in the cow shed. And so sometimes we blow out, and and the media seems to give numbers, but they're they're not referenced against anything and they you know they polarize people into opinion. Could there be the ability to create a policy that says, hey, you're you need to, if you're gonna do data centers in New Zealand, there needs to be coes loop water, you need to, you know, prove that that's the case, and you need to be responsible for all your power supply as in be net neutral in power consumption. And so these, you know, this business is getting more money thrown at it all over the world than any other. The cash is there, pay for it. Like pay for your energy in that it doesn't affect the other citizen who's not uh, you know, an owner of your data center. And so keep the the from a sustainability, keep your your thermal footprint down, closed loop water, just like a radiator on your car, it recycles, you don't keep putting water in your radiator, same type of thing. With just some simple legislation change, I think we could remove a lot of the angst that is being beaten up against um data centers. And then my other thing is what about we just create distributed data centers? Each one's the size of a 20-foot, 40-foot container. You dot them all over the place. You know, already Tesla is, I think they've given 10 million back to people using their Tesla power walls in the US to uh to offset you know grid consumption. All you'd be doing is doing the same type of thing. Let's put compute and and energy consumption at the edge, micro, and and it I it addresses other New Zealand's other biggest issue, which is our volcanic infrastructure, right? Is that you create this distributed network, no, you know, there's at least three copies of all data at any time. So if there's a a national disaster, you're not going to wipe out a data center that's in the wrong place at the wrong time. Is it are we just not doing enough thinking around this and actually coming up with you know, new world solutions for the new way the world's going?

Legislation vs. Self-Regulation in Data Centers

Oliver Hartwich 21:47

Yeah, I would agree with all of that. I mean, the only thing is do we even need legislation for that? Or wouldn't it actually be in the company's own best interest to go ahead and just do it? Because in the end, you want to keep social. license, you want to be welcomed in the country in which you invest. So if you are an investor looking at New Zealand from overseas, you would probably ask yourself, so what do I have to do in order to become an accepted part of the New Zealand fabric? And in which case, I think the obvious thing to do is actually to take care of your own water needs, to take care of your own energy needs. So you can come in with a positive story and explain to the New Zealand public, hey, we we are a new addition to your system. But by the way, there's nothing to be afraid of because we are basically providing for ourselves. Yes. And in that case, if you do that, I think you should have no problem convincing the New Zealand public that that's a good thing to do because you will of course contribute to jobs, you will contribute to economic activity. So what's not to love? So I think if you think through this properly, it might not even require legislation to do that.

AI Adoption in New Zealand's Small Businesses

Mark Smith 22:47

Yes, yes. Going back to the question I asked earlier about the US government turning off models and things like that. Is this going to create a scenario where open source open weight models get adopted by the rest of the world because they can have, you know, the biggest player in the space of course is China producing their tools like Kimik2 and a range, you know, Deep Seek, etc available that it would allow then a data let's say we had data centers in New Zealand, the most likely scenario if we can get the silicon to run it is that we're going to bring in these type of models into our own data centers, ground them on our own data, tweak them to New Zealand conditions, etc. Not worry about our data's going off offshore because we won't have a commercial relationship with the anthropics, with your open AIs or or or the other hyperscalers for that matter.

Oliver Hartwich 23:50

Sure. And the additional benefit of all of that would be that you are totally in control of your data. Which is something that of course if you're working in government might be of interest to you. You would not necessarily be comfortable with having your data stored elsewhere whether that's actually in the US or in China or wherever else. So in order to get the government to move towards a platform that they can use with trust actually that might be an attractive option.

Mark Smith 24:14

If you look if we can turn our eyes for a moment to small business in New Zealand, which I think makes up around 97% of all business in New Zealand. So it's a it's a big it's a big slice. What are you seeing? Are you seeing any adoption type metrics? Are you seeing more fear than positivity? What are you seeing happening in the New Zealand landscape when it comes to adopting AI that that really you know everyone's talked about productivity in AI and I'm thinking reinvention. Yep me too of your business model right and I'll give you a simple example

The Future of Accounting Software

Mark Smith 24:53

I've been a zero customer for 15 years and I use that SaaS solution and each year it seems to get more expensive but I'm a small business. I don't need any rocket science here all I just need is to meet my tax obligation, my IRD obligation and that's it, right? In the smallest amount of time. Yet just this month fees have gone up again. So I turned around and said okay after a lot of uh input from my CEO which happens to be my wife in the business a small business that we have is why can't we just replace it as an I'll build the software myself. And so but I'm like I'm not interested in building a piece of software that I have got to sell like accounting just does not spin my wheels in the slightest but I have a personal need. So I kicked off a project an open source project up on GitHub and I'm building um what I call the NZ ledger which is the kernel of any general ledger accounting that every system needs. Doesn't need anything fancy but I've made it CLI MCP API and agent first in its design principle as well as human first right so in you know so high level of audibility security and the inability for people to fiddle a book so to speak. And so I'm open sourcing this because I just see it as it's a core to any other piece of software. If you're in business and you've got an electrical company you're going to have the supply of electrical components the job projects etc you run but core accounting is the same core accounting you just plug this in anywhere. Is you know and it's called the SaaS apocalypse right that that with that NZR right now there's their stocks are tanking the CEO's selling their stocks like no tomorrow and when a CEO sells stocks I'm always like man they see the future of their business why are they selling their stocks? What what is you what's your observation of one SAS apocalypse that's that's been touted and the opportunity for small business because what I did is I didn't say hey I want to create a zero competitor. I always said I want to create an accounting system of the future how would we do it if we took all the experience all the legal requirements all the regulatory requirements

AI's Impact on Job Markets

Mark Smith 27:05

how would we create a system that took advantage of where the market is going and it's come up like I didn't even think of hey you need to create a core kernel that doesn't change for anybody. It's the same for everybody and then allow anybody to tack their unique scenario on top of it totally freely no no no no issue.

Oliver Hartwich 27:23

Yeah well the question is about how much small business uh already adopts and adapts and when you look at figures you see um 80% of New Zealand companies are now playing with AI in some form. But does that actually tell us that they're doing it right? Or does it tell us that they've just experimented and maybe concluded actually doesn't work for them or they're not getting the outcomes they want. I mean just saying that you've got an AI subscription somewhere is a bit like saying you've got a card for the library. But it doesn't make you a well read person. It gives you access but what you do with it is a completely different thing.

Mark Smith 27:56

Brilliant.

Oliver Hartwich 27:57

What I would say is actually that smaller companies are probably in a way easier for AI use to happen because at least they don't have these big structures and big kind of IT policies and confidentiality policies and whatever other cybersecurity policies that you find in the large companies. The large companies tend to be a lot more bureaucratic when it comes to these things for good reasons. Because they often deal with confidential data and you wouldn't want to have a free for all in a large company. Can't have it. But at the same time all of this bureaucracy then means that they are very slow to adapt and they might only have one system. So imagine you're a large law firm. You are dealing with very confidential data. You you wouldn't want to risk your client safety and privacy just for the sake of having um access to the latest model necessarily you want something that you can trust and so you build this one model perhaps even a custom made model for your company and that's the only access that you give your employees and your lawyers. Well that's fine because it keeps the data safe and that's your number one priority. At the same time it locks them out of anything else that might happen. So you build a custom model and then you see that Claude Coburg comes along and you see it ChatGPT work coming along or Perplexity computer coming along or whatever else it might be. And you can see this privately this is really exciting. Imagine what I could do except I can't because we have just got this IT policy we built our own customized GPT and we're stuck with that now. And so the world of AI basically passes you by and you can't do anything about it. Whereas if you are in a smaller company even if it's a law firm a small one you know what you're doing you know how careful you can be and you can actually probably experiment with a lot more freedom than the large competitors can. And so I think that is probably a problem for a lot of large companies. The larger the company the more likely they are to become another sort of bureaucracy when it comes to IT and so that will probably slow us down.

Mark Smith 29:52

But meanwhile for the smaller companies do they actually have the right kind of qualifications and insights and people to actually make the most of it I think most companies probably don't at least not yet but so so the the advantages for them but they don't have the wherewithal to to to capitalize on it.

Oliver Hartwich 30:12

Yeah and it takes leadership it takes tech geeks like myself leading companies to then introduce their colleagues and say well actually have you thought about that and I've seen it here I mean I was probably the geekiest of all our colleagues in the beginning and I was super enthusiastic about AI when it all really took off there was a bit of skepticism among colleagues. I think we have now all signed up to that and but it it was a process and I I think it took a bit of leadership.

Mark Smith 30:38

What's your take on the using AI as a scapegoat and and this is doesn't just apply to New Zealand this is a global thing whether it is CEOs saying they're downsizing their workforce because of AI and when you dig below the covers you find there's been bad decisions made but AI is the good scapegoat to point to and then we have a certain finance minister in recent times talk about

AI as a Scapegoat for Job Losses

Mark Smith 31:03

you know 8,000 jobs out of the government public sector being reduced or whatever the number was and AI's the reason and therefore it just creates this fear this I feel unnecessary fear from leadership that have a stage that have a microphone that have the public's attention through media is this doing damage that's going to put us behind for for for some time? Does it create a hostile the people that are worried for their jobs right there's employment right now?

Oliver Hartwich 31:35

Yeah. Well two answers to the question first of all when the Minister of Finance says she wants to shrink the public service she is right to do so even without AI because it's too big. The second part to the uh answer is actually in shrinking the public service she can of course make use of AI and get the public service to become more productive and also better better in the outputs that they deliver. So all of the stuff is fine. But on the broader question um is it a scapegoat for job losses? Well it depends on what companies do. When I look at our own company here we still have the same number of employees we had before but our output has gone up massively I haven't fired anyone because of AI use but we're just putting out more stuff and stuff I could not even dream of before. I can give you a practical example. In 2014 we produced a little book it was called New Zealand by numbers. The idea was a very simple one we would look at New Zealand across a hundred different indicators going from the demographics you know life expectancy family formation household size to crime statistics to economic statistics productivity statistics whatever about a hundred different data points and we would look at how New Zealand developed over the past 50 or 60 years. I thought it was a relative project at the time we wanted to just produce a graph per topic with maybe two or three paragraphs of commentary around it. Well that simple project took us about half a year and it kept the entire research team busy and you know you find data you find different data then you discuss which is the right statistic then you have to produce the graphs you have to produce the commentary then you discuss it with colleagues colleagues agree then the whole thing has to be produced. It took us as I said half a year and at the end of the project we said never again that was such a nightmare. Then Fable 5 comes out and I thought well that's a test run. So I the first test run was actually I was um waiting for my flight to Auckland from Wellington and I thought well Anthropic told us this is agentic so I tried it in chat in chat I just gave it the old PDF and I said well how about we produce a 2026 version of that gave it that command and by the time I landed in Auckland I checked my Claude app and it had produced about 25 pages. So not the full report but I could already see how agentically it approached this task and I thought oh that's great. So anyway I flew back to Wellington later that night and then as I got home I thought well okay now do this properly. So same kind of prompt but this time in the proper app on my PC at home um and about two and a half hours later and using my entire fable allocation for the week plus an extra 80 US dollars I had the first draft of the book. And of course there were still a few mistakes in there took a few iterations which I then of course worked through with GPT and perplexity and basically filtering it all out but altogether this entire thing was produced within a week. And it was better and more comprehensive than the 2014 version and we published it. And because that worked so well I thought well I now we're going to do a project that I always wanted to to do which I never dare to put to any of our colleagues because I knew it would be a nightmare let's compare New Zealand across all of these data points with 28 other countries around the world. Wow and that is a massive task I mean as long as you compare within the OECD you'll probably find the statistics. Once you venture beyond the OECD and you look at developing countries or you look at you know China, India, all of this world getting data is really hard. It's time consuming to do all that research but again it took about a week and then we produced that one too. So these are two publications we would have never ever done without AI yeah but they are fantastic. So our output has gone through the roof.

Mark Smith 35:23

We as in if possible we'll get the links and put them in the show notes uh for the listener. This is gonna be kind of off topic quite a bit but I'm just interested if you have any view. Argentina and their national debt and there seem to be elimination of their national debt in a very short amount of time as I understand it. What's your observation in there? Have you researched like looked into anything or just

Argentina's Economic Transformation and AI

Mark Smith 35:50

what's your thoughts?

Oliver Hartwich 35:52

Massively encouraging to see what's happening there. Argentina of course has a long long history of economic trouble going over a hundred years, isn't it? Over a hundred years. They did everything wrong in economic policy you could do wrong import substitution, nationalizations, price controls, any kind of economic policy folly you could think of Argentina did it. And so they went from one of the most prosperous countries on earth to one of the most troubled economies. And so what's happening there under Javier Millet is hugely encouraging turning this around and turning Argentina once again into a more market driven economy. But then again early days and I'm not a huge fan of everything that Millet does especially when it comes to international relations but so far at least on the economic front encouraging.

Mark Smith 36:46

In that of all countries in the world he's going for and I might not have the exact correct phrasing but legal status for AI companies as an an basic company that doesn't necessarily have human people in control of it. Like an agentic company he's starting to to go down those paths or make it a a place in the world where if you want to explore that side of things, you know agents would have full autonomy.

Oliver Hartwich 37:14

He's certainly a politician thinking creatively I think we need more of that.

Mark Smith 37:20

Interesting times. Oliver it's been awesome to have you on I always find your your your view of the landscape so insightful thank you and uh I appreciate it.

Oliver Hartwich 37:30

Thank you very much.

Mark Smith 37:43

You've been listening to the intelligence age a mic the intelligence age is already here. The question is what we choose to do with it