# AI Friction Removal: The Next Billion Dollar Opportunity

**Podcast:** Another Podcast
**Published:** 2026-09-20

## Transcript

Hi, I'm Tony Cameron Brown.
And I'm Benedict Evans.
And I was thinking about friction as a very short and sweet thing to think about.
Yeah.
And the sort of the entry point, the way I was kind of coming at this is that, as I keep saying, you know, AI now is like the internet in like 1997.
And you've got early adopter.
tech savvy geek use cases like walking around San Francisco with your laptop open running your agents.
And most people aren't builders and most people don't build things and think about ways that they could automate their lives.
But what are things that you would automate if you're a normal person or that could be automated for you?
Go back to thinking about the internet again.
And it struck me there's a sort of a, you know, One example of many in that would be think about buying plane tickets and what it took to buy plane tickets before the internet and how the internet completely changed that without anybody really sort of sitting and thinking this is an enormous problem I need somebody to build infrastructure for.
All of which is kind of a way of saying like what are the equivalents of it's an enormous pain to get a plane ticket or it's an enormous plane to get that particular SCU before the internet.
How do I choose a credit card before the internet?
And I thought an example now would be to think about friction and to think about all of the places where theoretically, you know, you could save some money doing that, but it's an enormous pain and it's aggravation and hassle and you can't be bothered.
Yeah.
And sometimes that's many small things.
Like if you actually went through and optimized all of those every day, then you'd save a certain amount of money.
Sometimes it's surprisingly big things like your health insurance bill where you might, in America, you have this thing called health insurance.
And you might save hundreds of dollars a year if you actually sat down and carefully planned which one you should be using.
And there's all sorts of fields, entire fields of economics and all sorts of economics papers looking at all of these.
problems of friction and inertia and coordination, where in principle, if you had perfect information and you're willing to do the work, you could save meaningful amounts of money.
But in practice, no one actually does, or most people don't actually do this.
And you get that goes right the way through to extreme cases like these people who run complicated spreadsheets, optimizing all of their credit card rewards, where, you know, if you have an American, you know, these American credit cards, again, you know, they have that people call them coupon books.
So you can get $5 off DoorDash every three weeks and you can get $15 off this order if on Nike, but only on the first Thursday of every second month and so on.
And a lot of these rely on friction and rely on people not doing them.
Or maybe they don't, but anyway, people don't do them.
And so I thought, you know, there's a sort of a whole broad category here where you can say like, What happens if I can say to my agent or I can say to my bank?
And this is another interesting question.
Where does this sit?
What's the right integration layer?
What happens if the agent on my phone or my bank or my credit card, I can go and say, you know, go and look at all of these bills and go and work out where I can save money.
What should health insurance should I be using?
How could I optimize those?
And that is rather like the margin in plane tickets, not trillions of dollars in aggregate and not thousands of dollars for individual consumers.
But it all kinds of adds up.
So I was thinking about this the other day and there's a paper looking at just the American, just the California health insurance marketplace that if everybody in California was on the correct marketplace plan, that would be about a billion dollars less money going to the insurance industry.
So multiply that by 25 for the rest of America and then multiply that by 20 or 50 other industries.
And somehow or other, you've got 100, 200 billion dollars, maybe 500 billion dollars.
Go look at all of those credit card rewards and think about how those move around and get allocated.
So the.
Where is it?
The US credit card reward industry.
Let me find where is that chart?
Rewards earned.
Rewards earned in the US are 40 to 50 billion dollars a year paid out to consumers, which, of course, as most people know, comes out of the swipe fees, the interchange fee from the retailer.
How are those allocated?
What's the most efficient way for people to use those?
Where do they go?
How do you use your points?
You and I both fly a lot and we have these absurd balances of air miles and then we look at them and you think, Christ, how on earth am I going to use these?
Where am I going to actually allocate the right place to use them across the travel that I have planned, across the rewards availability, across the routing?
And you go to places like NerdWallet or the Points Guide, people go into dire businesses.
out of running this.
And it's sort of interesting, as I said, to kind of map that and think about, well, what did the internet do with like information liquidity?
What does AI do for kind of friction, the removal of friction?
Sorry, there's an obvious mixed metaphor there.
The extreme case is, as a bunch of people have said, that the marginal cost of arguing goes to zero.
So there have been a bunch of stories about, like, I think the UK employment tribunals.
So, you know, you're unfair dismissal, you can go to law.
And these tribunals are sort of overwhelmed with people sending in 200 page submissions where previously they might have written a five page letter because people are just using AI to generate legal paperwork.
And there's a sort of obviously there's a narrow point, which is to say maybe a lot of these get the law wrong.
But that's kind of not really the interesting question.
The deeper question is, and of course, the tribunals in turn will end up using AI to read all of these things.
So you'll have AI talking to AI.
But how much does that, again, flatten the playing field?
How much does that open up access to law to people where it was too hard, too expensive, too difficult to understand to do it?
And it's interesting thinking about this because I initially thought...
okay, we have access to information and better information.
that we probably didn't have before.
But the more I'm listening to you talk and the more I've been thinking about it is it's not actually access to info, but it's also the access to the staff, quote unquote, which is what you always talk about.
Like it's you all of a sudden have all of these interns that you didn't have before.
It is the, you know, you now can have an accountant, you can have an agent, you can have a lawyer, you can have a fixer, all things that actually we weren't privy to previously as normal human beings.
So it's interesting because I think I've flipped my thinking of...
oh, wait, it's access to, we're democratizing access to intelligence and information to actually, you know, it's headcount, which feels interesting.
It's boredom.
Yeah.
The boredom threshold and the aggregation and the having the right kind of mind to understand this stuff.
So, you know, I remember when I lived in San Francisco and had to go through the process for applying for a kindergarten for my son.
And it was this horrible, complicated form.
And, you know, I had to sit down a couple of hours and sit and scratch my head and go and do a bunch of research and go and find out about what all these schools were and where they were and understand the process.
And, you know, you know, this is kind of what I do for a living.
And I found it like a nightmare, frankly.
And, you know, then I remember, you know, we got into a school that wasn't the school that we wanted to be in.
And this is a bizarre lottery system.
It was a school that was on the other side of San Francisco, but you could appeal.
So we went to the school to appeal.
And there were a bunch of people there who'd clearly like this was their first encounter with any of this and had no idea how any of this worked.
You know, this is always the challenges that bureaucracy is a regressive tax.
that to say bureaucracy privileges people whose education and background and experience is about managing complexity and understanding how to manage complexity and it penalises people whose experience and expertise isn't that.
You know, bureaucracy is a tax on the poor for the benefit of the middle class.
Same point, incidentally, about credit card rewards.
There's an analysis that says credit card rewards in America are effectively a wealth transfer from poor people to rich people because poor people use debit cards where they don't get credit card rewards and the prices are pushed up to pay for the credit card rewards that are going to rich people who are paying for their $500 a month credit card or $500 a year credit card.
That's kind of go down a rabbit hole.
But there's a sort of general point here about access to, you know, is to go back to my thinking about what the Internet did to access to information, that you didn't need to be live in New York or London or L.A.
or Chicago and have access to that library and know that library was there and go have the time to go to the library and spend a day finding it.
Now you can just open Google and you have the thing 10 minutes later.
And that's what I mean.
Like, it feels like it's not just access to info, it's access to the headcount to actually do it, even if the headcount is your agent.
Yes, if the internet was access to the information, this is access to, is it the librarian?
Is it the expert?
Is it the person who could sit and do the thing for you, who could actually understand?
The internet was, now I can...
you open up knowing where the thing is and how to get it.
And the new one is now you open up knowing how to fill in that 50 page form and how to get it.
Your boredom example is perfect of just like, OK, now we know we have all the information, but you still have to go through it and read it.
And it is just not an enjoyable experience whatsoever.
Whereas now it feels like, oh, wait, you can actually take that away as well.
And yeah, boredom may be too dismissive a term, but, you know.
There's a bunch of people, the large portion of the population where they could spend, you know, two days on this and they'd save an amount of money that is meaningful, but not meaningful enough for them to spend two days on it.
But if it was five minutes and pressing a button three times, then a lot more people would do it.
It also almost feels like we've been promised this sort of...
ease and access before just like comparison sites and you look at everything that we've just talked about that is tied more to the access to information.
I wonder what is structurally different now that means that this could truly really work and why it might feel really different of and maybe it goes back to again it's the headcount piece it's not we don't just have the information at our fingertips but we can actually get someone else.
Well an obvious answer is that an obvious answer is that you'll get back to where you were before.
I mean, you know the joke that half of AI will be turning three bullet points into an email and the other half is turning an email into three bullet points.
And that's obviously the challenge with all of these individual legal filings.
You can certainly make an argument that if AI makes complexity intelligible, the result is an explosion in complexity.
And so you could have massively more granularity.
So in fields where there's no marginal cost to granularity, and the reason you didn't have that granularity was because it would be too complex, A, for you to manage and B, for your customers to manage, you could hypothesize that AI might result in an explosion in granularity in some fields.
You can also hypothesize that it results in attempting to move friction to other places.
You could hypothesize, again, as we've talked about a lot, there's a sort of the appeal of analog.
the appeal of, of, of, of authenticity, that if this stuff can be automated, um, where does that move the value to the stuff that you can't automate?
Um, part of this, I think is if you, if you think about like the, the experience of, of, of retail banking, it used to be that your, your, your, you had a bank and a branch and that was your branch.
I don't know if this happened, how much this happened in America, certainly in Britain is when I was a.
you know, young adult, that was my branch.
Same in Brussels.
You'd have your person.
They would ask, which branch is your branch?
But I didn't actually have a designated person there, but I was, you know, I had to, you know, I was, I had an account at that bank with that branch.
I remember HSBC in London going to one and they're like this and walking in and going, we can't help you.
This is not your branch.
Please go.
And I'll be like, wait, what?
But that's a sort of a legacy thing.
The reason I mentioned this is that it used to be that with a before computers, that branch had your stuff.
And that branch was the place where they could evaluate you and decide whether to give you a mortgage.
And it was your bank manager.
There is no such thing as a bank manager now.
There might be somebody who like notionally manages that branch, but they're not a bank manager any more than like the head of the Starbucks is a Starbucks manager.
They are just, you know, a retail supervisor because all of that got computerized.
And so there's a layer there of, you know, what.
decisions required people and couldn't be done because they required people or that you didn't do because they required work.
Again, I'm slightly thinking out loud here, but again, the analog I'm thinking of here is the internet, what the internet did to information.
This does to a certain kind of work, a certain kind of admin and hassle and paperwork.
and aggravation and friction, that it brings liquidity and transparency and removes friction from that in the way that the internet did for access to certain kinds of information.
Would you say then that the last sort of remaining friction that we are faced with is legal?
You mentioned just like your agents can figure out what the right mortgage for you, but they probably can't sign it.
So does that become something interesting of just that we talk about we're writing all of these legal documents.
I don't know.
It just feels like Legal feels like the last friction that we can't actually solve for right now.
And the importance moves to people who can actually sign the paper, notarize.
Well, this is the thing that people in tech in Germany complain about, that you've got to have a notary, you sign a contract, and you've got to have a notary sit with you and read the entire thing out to you.
So you've got to...
You incorporate a company, you've got to pay a notary to spend two days reading out the articles of association.
You sell your company, you've got to pay a notary to spend a day reading out the terms, even though you've read them and you've paid a lawyer to read them, you've got to pay a notary to read them to you.
That almost feels like that's the last friction that we're faced with.
Yes, these are deliberately restrictive practices, like the analogue whole.
Maybe that's like the...
you know, the last moment of protection.
Like, you know, if you can hack the security cameras, then you'll have to have a human guard there.
You'll have to do a human pat down if you can hack the security cameras.
But that, you've got a little bit of this now of like when you do a wire transfer, they want to call you.
And you think, okay, it's not very long until that's an easily, that can be hacked too.
And I can do a video of Tony saying, yes, I authorize this wire transfer.
I don't want to milk the subject too long.
I just think there's a broader point here, which is to go back and look at, if we go back and think about, if you'd been having this conversation about the internet in like 95, 6, 7, 8, or about PCs in the early 80s, you could make very, very high level statements about what this means.
Like, you know, information wants to be free and there'll be no borders.
And, you know.
you know, physical distribution, the value of physical distribution disappears and what have you.
And then that's the aggregate of innumerable industries.
This is the quote from Carlisle, history is the aggregate of innumerable biographies, which is not the only way of thinking about it, not necessarily the best way of thinking about it, but it's a useful way of thinking about it, that these vast technology changes from the internet were the aggregate of innumerable individual things.
that were being done without the internet and are being done with the internet and what did that change?
The conversation I was on immediately before this, I was talking with a big industrial company and the framing I'm thinking about, one of the framings I'm thinking about as I work on my new presentation for the autumn, is that for any industry when the new thing arrives, you're in one of three categories.
you might be in the category of this is purely useful at the margin for operations.
So like if you're a mining company, how much did the internet change?
Well, it was useful.
And you could email that 500-page report instead of FedExing it.
But at the end of the day, your business is digging holes in the ground and carrying trucks of sand and selling it.
And so actually it didn't change the nature of your business.
It was just a useful tool around the margin.
And it probably didn't actually change your margins, your profitability either.
It was just kind of useful.
Then there's like a middle case, which would be to think I was thinking of examples of like airlines or banks, where this changes an awful lot about how you go to market and how you present yourself to your customers.
And it creates new competitive opportunities.
And for the sake of argument, the internet made it much easier to be a budget carrier, maybe.
Maybe, maybe not.
I don't know.
And it's a new entry pass for banks.
So like Revolut can get hundreds of millions of customers without needing physical branches.
But actually you bank with HSBC and I bank with Chase.
And it's a new competitive vector, but it doesn't just make the entire industry change.
It doesn't demolish any incumbents.
And then there's a third case, which is like what the internet intermedia companies say, where it actually does completely demolish everything.
And often you're not clear if you're in category one or two or if you're in category two or three.
You generally know if that you're not in category one or three.
That's kind of an easy question.
But you don't know if it's one or two or if it's two or three.
But that then is hundreds of industries.
And the impact of the internet on taxis, Uber on taxis was very different from the impact of Airbnb on hotels.
This is like the great Marc Andreessen line is, you know, Uber doesn't sell software to taxi companies and Airbnb doesn't sell software to hotels.
But Uber demolished the taxi business in a bunch of cities and Airbnb did not demolish the hotel business.
It was like a new revenue stream that's like five or 10% additive.
All of which is kind of a slightly long-winded way of saying, like, there's going to be lots and lots and lots of those places where AI goes and does this one thing.
And that might completely change the industry or it might not.
And those kinds of...
Or just change where the revenue is coming from and where the revenue is going.
Yeah.
And those kinds of marginal frictions where we just threw out half a dozen examples, and I'm sure people listening to this can think of half a dozen more.
where the AI will just automate away that thing that was kind of a pain and you won't have to do that anymore.
And then you'll kind of forget that there was a time that you had to do that.
And that was like $2 billion across America.
And the last thing that I would say there is where it becomes exciting for me is what are the things that it will be able to automate and do that I don't have to go and verify.
I don't have to run for everything and check that it's correct and accurate because that's where I think it loses.
momentum for me or it loses excitement or it loses its appeal.
It's just like, well, you did all the hard work.
Great.
But I feel like I need to spend another two hours just triple checking everything because I'm not 100% confident that this is correct.
Yeah.
Well, that becomes a question of finding the right way, finding the right, mapping that against the right use cases, finding the right ways of presenting it, you know.
These are probabilistic systems.
Sometimes you can rely on them.
Mostly you can't rely on them to be 100% correct.
So how do you build a product around that?
Which, again, is a challenge of how far up to the stack of the models go.
Many of the examples we've sort of thought about, yeah, you could imagine that a general purpose model could do that.
Many of them you also think, no, I want tooling.
And I want GUI and I want structure and I need a company and I need testing around that to make sure that it's being done in the right way.
But again, this is my Uber versus Airbnb example.
Like, it depends.
It's a good place to end.
There we are.
21 minutes.
I think there's a union rule that you're not allowed to do a podcast in 21 minutes.
It's just like, it's one of those ways, like, no, no monetization.
210 minutes.
21 minutes.
What is this crap?
Yeah.
I know, add a zero to our time count.
But I think that's kind of, that's just one piece to pick up and think about.
And there will be a billion dollar company in what I've just said, probably at some point in the next five or 10 years.
Amazing.
Talk to you soon.
There we are.
Talk to you later.
Bye.