269. Monzo founder: AI will kill income tax

0:01Humans are not economically useful anymore. I think income tax will go away pretty soon. No. Being a senior lawyer or a senior doctor will seem like a joke. We started in 2015 and by 2019 we'd grown from a team of 10 to a team of 2 ,000. This is Monzo we're talking about now. Yeah, this is Monzo. We were losing about 100 million a year under attack from the press. You were at Starling, you fell out with your partner Anne Bowden. There were 14 people at Starling and 13 of us started Monzo. And you haven't spoken to her since?

0:40hello and welcome to the rest is money with me robert peston no steph today which presumably means she's out fighting crime again um today i will be talking to tom blomfield he is one of the most successful entrepreneurs that britain has produced over the last 20 odd years as a founder and creator of Monzo. He now is a partner at Y Combinator, which is one of Silicon Valley's most successful incubators of tech businesses. The beginning of our conversation is going to focus on this AI industrial revolution we are living through because his views in terms of how all of us have to adapt, but in particular, governments have now got to massively think about how they fund the welfare state and how they provide decent incomes to people as we go through this massive jobs upheaval what he has to say is very challenging so here is my interview with tom blomfield so i want to start actually with where we are in terms of how ai is changing working and business practices um as somebody who is really at the sharp end of sponsoring young tech businesses.

1:57Just talk me through how big a moment you feel this is. I struggle to find words to answer that. And this is going to sound hyperbolic. I think this is undoubtedly a sort of technological and industrial revolution on par with anything we have ever seen in human history. And as the words leave my mouth, I can hear your sort of audience sighing and shrugging in the background in the UK. And I'm giving you this story from Silicon Valley, the breakthroughs that have happened and will continue to happen over the next months and years feel easily bigger than the internet. And I think will reshape the economy and society.

2:38It's quite unprecedented. So one of the things that I have experienced, because I've obviously not with sort of granular the details that you have, but I've been observing these astonishing technological changes that we'll have, are already having profound economic and social changes. And I've been talking about it very loudly, very publicly for two or three years now. And one of the things, though, that I observe is whenever I say the kind of things you just say, people sort of, I don't exactly glaze over, but their brains freeze and they try and change the subject. It's it's it's. So, for example, I mean, I've talked to incredibly senior people in the British government about this, and I've been doing it now for about three years and they don't want to hear.

3:30They genuinely don't want to hear. So what would be really helpful is just give us two or three recent practical examples of what's going on in in your world that maybe you know those for whom this all feels too big can sort of connect with if you see what i mean we're talking basically about large language models about these um very expensive ai models that are trained on uh basically all of the data that they can get their hands on all of humanity's recorded data that's books and the internet. And they're now paying experts to produce more data at the frontier of these subjects. And the output is very generalized intelligence that is able to do narrow human tasks currently better than any human, more or less.

4:19I think we're at that level where for narrow domains, these AI tools are performing beyond PhD level, beyond university professor They are actually beating humans in narrow domains. They're not yet generalizable. When you cross domains, they might seem a bit stupid, if that makes sense. So they're sort of very narrow geniuses. But the progress is going so quickly that by the end of this year, I think by the end of 2026, they will be generalizable. And a specific domain that I'm very close to is coding. A year ago, Dario, the CEO of Anthropic and Sam and other people very close to the industry, leading the industry, made warnings like in software engineering, humans are not going to be writing code in a year.

5:12And they were basically laughed at. Like a lot of software engineers said that's impossible and never happened. And it's happened. There was a big leap forward in the coding models in about late November, early December of last year. So it was only four months ago. And for people who are on the cutting edge of this, they are no longer writing code. And that doesn't mean all software engineers worldwide are no longer writing code. In fact, probably 98 % of them are not using these tools to their full extent. But the people who are living at the edge, the cutting edge, are not writing code. So I work at Y Combinator, an early stage investor.

5:44We did a survey yesterday of 200 of our founders who we just invested in. So very, very early stage companies. Very small, you know, a lot of caveats. small codebases. This is not a 10-year-old legacy codebase, but our founders, out of 200, there were about five or 10 who said that the AI is only writing 75 % of their code, and the remainder said the AI is writing 99 % or 100 % of their code now. That was not true a year ago. Tom, so much more I need to ask you. We'll be back with you after a quick break.

6:21we're proud to say that the rest is money is powered by octopus energy this year uh greg jackson is back to answer another question now this is something that we see a lot that i wanted to ask you about when you start in something new how do you think about risk and momentum i think people think entrepreneurs love risk but i'm not sure they do i hate it i never buy individual stocks and shares on the stock market. I don't gamble. I think the thing about an entrepreneur for me is I've got more control. When you're working for a company, every day, you're at risk of what that company chooses to do with you.

6:56If you're an entrepreneur, actually, you've got far more say in what happens tomorrow and next year than when you're, you know, at the mercy of your bosses. Nice one, Greg. Well, thanks to Octopus Energy for powering this episode of The Rest is Money.

7:13So is there any scenario where there will be significant numbers of humans employed in writing code over the next, let's say, 18 months out, something like that? No. But I think those humans will still be useful, but I think their jobs will change very dramatically. I don't think we will – there'll be holdouts, there'll be legacy industries where they can't dot these tools, banks and nuclear power plants, perhaps, where they will still insist the humans write and review every line of code. But in 18 months, no, I don't think humans will be economically useful in writing code. I still think those humans who used to write code will be employed to solve business problems.

7:57They'll talk to non-technical people and translate the requirements into, you know, they'll instruct the AI to build the systems and they will still be useful and productive. And I think the demand for code will go up dramatically. I think this is a really important concept, sort of the elasticity of demand. So as the cost to produce code goes down, the demand for it, I think, will skyrocket. I hate any phrase with vibe in it, but it's what's known as vibe coding. It's where basically we instruct something like Cowork to do all the code, which is Anthropics service. We instruct something like Cowork to do the coding for us.

8:38Yes. And for our personal use or primarily for our business use, actually, I think many people who will go to their job and say, I used to do this thing repetitively. It's very tedious. I will build myself a quick little program to automate most of my job. And so the elasticity of demand in software, I think, is very, very high. As the price goes down, the demand for software goes dramatically higher. One of the things I'm obviously fascinated by are the economic implications of all of this. I think I'm currently paying Claude for the sort of medium tier service. I think I'm paying them something like£180 a year, something like that.

9:17I mean, just on a personal level, that's going to put us, you know, if I happen to want to employ a coder, I'm not going to be paying more than£180, am I? So the implications for anybody who is still a human coder in terms of what they might be able to earn are pretty depressing. I think for maybe the median coder or like the bottom 50%, it's very bad. I think for the top coders, I think they just become dramatically more productive. So I work with Gary Tan at YC. He's been coding for the last, more or less constantly for the last 60 days and produced a piece of work that he estimates would have taken him about five years previously.

10:00And he has spent about$40 ,000 now on coding tokens, on paying Anthropoc and OpenAI for their coding tools in about a month or two. So, Gary, who is a good coder, plus these tools, is able to do the work of a software team over five years, but deliver it in a month or two. So, there's the productivity, which is incredible, but also the savings. We're talking about potentially savings of millions of dollars there. Absolutely, yes. All right. So I'm now going to put to you your own, you know, I don't know whether you felt it was a sort of bit of a comedy classic when it happened, but you did a bit of vibe coding.

10:42You got a bit of attention for, I think it was called Recipe Ninja. Quite a lot of people poked fun at you when they were able to make, I think it was cyanide ice cream or, and then there were some, there were some one or two. So there were one or two other recipes, which for a family audience, I probably won't repeat now. Now, I think you thought, and I absolutely get why, given that you were able to fix these errors very fast, that actually you'd proved your point. But what did you learn, as it were, from this experience? I mean, it was a toy project. This was not a company that I was trying to build.

11:17It was not an economic endeavour. I built this over about a week or two, and this was a year ago. So this was really before the great breakthroughs we've had recently. Yeah, but still there were breakthroughs. So I used these tools in about two weeks to build a fully voice controlled recipe app with AI built in just to see if I could. And before AI, I would not have been able to build this. I'm an OK coder. I'm a reasonably good engineer. But this would have taken me. I don't think I'd have been able to do it. It required so many different technologies I wasn't familiar with. I simply wouldn't have been able to do it.

11:50And in about two weeks, I was able to put this site online. and of course people have joked around with it and who cares, it was a side project. But it was eye-opening for me because I was able to produce this thing that was so dramatically beyond what I ever could have done before. And that was a year ago. And my guess is if I did that today, I could do it in one hour. And I sort of proved that to myself because I've been building a project for the last week or so, which is basically a new form of Google Docs with AI built in, sort of AI native. document editor and book writer. I'm in the middle of writing a book.

12:27A mountain quality of software I can now build today, even compared to a year ago, is an order of magnitude higher, which in itself was beyond anything I could have done manually. And it just left me sort of astounded, honestly. I mean, one of the things, though, that is obviously also hugely important is not just how AI is replacing this important but narrow task coding, but as somebody now who has been playing around with co-work for a few weeks, there are all sorts of professions where I am genuinely wondering Whether in a year or two, there will indeed be opportunities to go in at the ground floor, because so much of what junior people in, let's say, accounting or consulting or law are paid to do can now be done by an agent that more or less anybody can build and instruct.

13:31It feels, as you said at the beginning, like a very profound moment of job displacement for the kinds of professions that we regarded as pretty secure, you know, white collar, middle class jobs. I fully agree. I think coding is just the canary in the coal mine. The good thing for coding is I think there is huge elasticity of demand. And I think coding is actually the way you automate all of these other jobs, if that makes sense. So by being able to, because the AI can write software, it can go into a legal job or an accounting job and write its own software to automate those jobs as well. But I think if you think about the elasticity of demand in, let's say, tax accounting, let's say tax accounting becomes 10 times cheaper.

14:22Do people want to do 100 times more tax accounting? No, they don't. Once their accounts are done, they're done. So I think there are just some professions will get dramatically compressed and almost no humans will be required at all. And I think, as you say, the implications for society will be enormous. are enormous the one sort of i don't know sort of paradox in all of this and i just wondered if you've given any thought to to it because it is something that i'm sort of quite exercised at the moment is i do think that experience human experience still counts for something and so if If I had an incredibly complex legal or financial challenge, I would definitely trust the AI to do the original sort of the basic groundwork for me.

15:15But I almost certainly would want to have a conversation with a senior partner or just somebody who'd been in this space for decades, know the sort of some of the decision makers and the pitfalls and all of that. And so I think one of the sort of almost tragedies for young people today is, you know, quite difficult to see these firms hiring the same numbers of young people to rise through the ranks. But equally, there are still things that I think humans will continue to sort of, in a sense, be better at is not quite the right phrase, but will have an advantage in. But it requires sort of decades of working and accumulated knowledge.

16:09A question I'm really posing is, if all these jobs are wiped out, how do we get the humans who acquire the decades of knowledge who will nonetheless be useful, if you see what I mean? Because once those people are dead, they're dead. I think there are three or four things in there that I kind of want to unpack. I agree that the professional service firm used to be a pyramid and will go to a structure that's more like a pillar. I don't think you need so many junior people. I think for now, the senior people on very complex subjects are people will expect that. I think that's going to change in a year or two.

16:46I don't think it will take 10 years. So does that mean that you think you'll be out of a job? I think we'll all be out of jobs. Don't blind me. Okay, that's quite a term. I don't think humans will be economically useful in the medium term. And I think putting the I think that that claim is the only hard part about that claim is what is that time frame? So I think knowledge work, I think we will be very surprised very, very soon at how bad humans are at knowledge work, actually. Like, I don't think we are. I think we're basically primates who are evolved to, like, make very quick gut based decisions and deep analysis.

17:26we are, I think we will realise we're extraordinarily bad at deep analysis, thinking about second and third order impacts. And I think when the super intelligent AI comes around and it will come around very quickly, it will be laughable to think back and assume that we were like the smartest thing on the planet. You're basically saying, although we are very good at, you know, heuristics, rules of thumb, you know, we learn by experience and then we apply what we've learned. almost automatically a lot of the time. Broadly, what you're saying is the machine will do that, though, with a level of sophistication and foresight that we're incapable of.

18:07Yes, and the ability to ingest lifetime's worth of human information at once and reason over it in a way that is superhuman and come to conclusions that we are just incapable of. And in that world, being a senior lawyer or a senior doctor will seem like a joke. I think it'll take humans a little while to adapt to that, but we're seeing it already in medicine, where you go to a medical professional and get a piece of advice and then go to chat GPT and it solves the thing that the human couldn't. And I think once you've experienced that three or four times, you're not going to go back to the doctor or the lawyer anymore.

18:40And so I think knowledge work is the first thing that falls. I don't think humans are economically useful for knowledge work in the future, not right now. And I think that timeframe is probably four to five years, maybe. And then I think humans will be economically valuable for moving physical things around still, you know, being an electrician, a plumber and a bricklayer until robotics come. And robotics is probably on a seven to 10 year time frame. In my mind, the rollout will be slower because you need to manufacture the things and it's, you know, atoms are harder than bits. But I think once you have super intelligent AI and pervasive robotics, humans are not economically useful anymore.

19:19And I think that will happen in a lifetime. And I think that will upend the social and economic structures of at least the Western world. How do we decide who?

19:32I think this is an easy prediction. I think income tax will go away pretty soon. It's sort of five to six year time horizon. I don't think we will tax human labor. I think we will tax compute. I think we'll tax data centers, inference, tokens, whatever you want to call it. And then we will use the proceeds of that to pay for stuff, to pay for the government. Some people talk about the world that you describe as the world of plenty, where the marginal cost of production is basically nil. The problem, as you have very astutely pointed out, with the world of plenty is in that world, nobody is working for money.

20:10So, you know, essentially the funding for the welfare state just vanishes, you know, goes down to zero. and therefore you've got to be clinically insane not to take what you say seriously. You know, we should be, governments should be actively involved now in debating and preparing for how they extract the income that means that, you know, public services, health services, schools and all the rest of it are funded and also that people don't starve because, You know, it's in the end, it'll be a world where there will have to be transfers, essentially of income from the machines to people. Otherwise, we all starve.

20:57How do we make essentially elected governments sit up and take notice that even if right, even if this is a so-called long tail risk that doesn't materialize the history of long tail? I mean, one of the things I keep saying to these people is you didn't think there was going to be a global pandemic. There was you didn't think there was going to be a global financial crisis in 2007, eight. there was we you know you didn't think global you know climate change was real it is now you know at what point you know one says you know one says to elected people when people are saying to you this is a this is a clear and present danger and even if you don't believe yourself it's a more than 50 probability you know you are literally insane not to be preparing for these outcomes So how do we get people to sit up and take notice?

21:52So I think the first problem is the time horizons are probably just over the edge of the election cycle. So this is terrible, but it's probably the next government's problem, not this government. But hoping we have elected officials who care about the future of the country, not just their tenure in parliament. I think the thing that I would push very heavily for is domestic data centers. I think at the moment, the world leading models are coming out of Silicon Valley and China. They're really the only two credible people who have these frontier models. There's some open source stuff. Taxing a Silicon Valley company to pay for British welfare state, I think is difficult.

22:37But I think if you have domestic data centers, it becomes much easier to say, we are taxing the compute or the tokens or the inference, whatever you want to call it, in these domestic data centers in the UK. And it's not too late, in your view, because one of the things I constantly hear from both people in your world and indeed people in government is we've just lost that race. No, it's not too late. The build out of data centers is happening right now and it's slow everywhere. and this is a question of planning permission and political will, but we need to 100x the building of data centres in the UK and then to support that we need to dramatically increase energy production in every way we can, nuclear and renewables and whatever we can do because these things draw quite a lot of energy.

23:27And so I think if we can do those two things very quickly over the next five years, I think we might be in quite a reasonable spot to extract the compute tax to then pay for all this, you know, the people who need supporting. But we need to act now. I'm sorry just to sort of ask the dumb question, right? But let's just say we do, you know, build a sort of so-called sovereign compute and data center capability. What is to stop, you know, Dario Amadei at Anthropic or ChatGPT or, you know, Alphabet and Gemini from simply running all their AI services, as it were, through, you know, data centers that Elon Musk will put into orbit around the Earth?

24:21It'll be extraterrestrial, not just extraterritorial, as it were. So what the hell do we tax in a world where, even though we built the data centers, nobody uses them because they don't want to be taxed? Yeah, I think the data centers in space one is a harder one. I'm an investor in StarCloud that just raised a very large Series A to put data centers in space. I think this is going to happen. I think the timescales are probably five to seven years before they're truly like industrial scale in space. And yeah, at that point, you just try to have I think you governments have to pass a law, which is if you are providing services in our country, you must pay the tax on the compute, wherever the compute lives, whether it's in space or in the Sahara or whatever.

25:07If essentially we're being replaced by, you know, AI powered and they're all robots in different, you know, some of them are physical robots. Some of them are dematerialized robots, but it's all basically robots, as it were. You know, why is it just impossible to essentially tax the robots, tax the services, you know, in the country in which they're used? I mean, in the end, that's presumably what you've got to do, isn't it? I mean, you've got to attribute some revenue. And you've just got to attribute the tax. You know, the HMRC will have to attribute the revenue that's being generated and then say X percent of that.

25:42I don't know, 40 percent or something has to accrue to the British government. Otherwise, you know, you can't do it. We go back to being manual labor. Yes. But if you look at our track record of trying to do that to companies like Meta and Amazon and Google, we have been spectacularly unsuccessful at that so far. No, no. Look, I agree the presidents aren't great, but if we don't start working on it now, and presumably in partnership with the rest of the European Union, it would help a bit. Because it does also seem to me that one of the lessons of this kind of global, these sort of global changes, is you do need cross-border cooperation between different governments.

26:17The more the governments are united on this, the harder it is for the providers to dodge tax unless they move to Mars. I agree with you, but I think we would have more political leverage to levy those taxes if the companies had substantial data center installations in the UK. In the negotiation, I think we just have more arms to twist if they're really running a lot of compute out of the UK. The thing we can do as a nation is authorize the construction. We don't even have to pay for it, right? We're like, there's enough money out there. There's enough people want to build them. We just have to identify the site and say, you may build data centers here.

26:56And we probably have to authorize the building of a bunch of power plants to power them. If we do those two things, I think they will be built. We just need to compute capacity. Even if they're running American models, I don't think it matters. I don't think it's realistic that we have a British frontier model. And I think, honestly, open source will also be a big part of this. So right now, the open source models are six to 12 months behind the frontier models. And I think what we will see is the frontier models will accelerate through to superintelligence. But we actually don't need superintelligence for most of the work we're doing.

27:33Normal clerical work probably needs average intelligence. and the open source models will provide average intelligence and we can host those open source models in our data centres and we can tax them and we can get revenue for the government and we'll be happy. Tom, there's tons more I need to ask you in particular about the business and indeed mental health challenges you faced when you were creating Monzo and we'll pick all that up after the break.