Your energy bill arrives. The number is bigger than last month. You cannot tell why. Used more? Unit rate up? Estimated reading? The bill does not really say. You squint, mutter, file it under “deal with this later.” Later never comes.
This time you take a photograph and say: deal with this.
By the time the kettle boils, it is done. Your usage has been compared to last year. Every tariff in your postcode has been checked. The exit fees have been read. A better deal has been picked and the switch has started. You saved one hundred and forty pounds. The agent took twelve percent. Receipt in your inbox.
This is not science fiction. The pieces to do exactly this exist now. What was missing was a way for narrow, useful work to pay its way. That changed in the last twelve months. Most people, including most of the people building it, have not yet worked out what it means.
The internet you grew up with
The commercial logic of the internet for the last thirty years ran on three things. Adverts. Software licences. Attention.
Adverts were the largest. Google. Facebook. YouTube. Instagram. TikTok. The biggest companies of the modern era made their money by capturing your attention and selling it to advertisers. The product was not the service you used. The product was you.
Software licences were the second. Microsoft, Adobe, Salesforce, the entire enterprise software industry. Pay a fee for the right to use a tool. Pay it monthly. Pay it whether you used the tool or not.
Attention was the underlying resource that made both work. Every interface decision, every notification, every recommendation, every infinite scroll, was tuned against a single metric. Time on platform.
This was not anyone’s evil plan. It was the rational response to a constraint. Building useful software was expensive. A working product needed a team, a year, several million pounds. The only way to recover that cost was to either licence the software at high prices to large customers, or to give it away free and recover the cost through ads. Both routes demanded scale measured in millions.
So the only businesses that got built were the ones large enough to justify the cost. Everything else simply did not get built. Not because the problems were unimportant. Because the unit economics did not work.
We got social networks for billions of people. We did not get coordination tools for the volunteer fire service. We got news feeds optimised for outrage. We did not get decision support for rare medical conditions. We got ten thousand variations on the same online retail experience. We did not get tools tuned to specific learning differences in eight year olds.
That is the world we have been living inside.
It is also the world that is now ending.
What changed
Two things shifted in the last eighteen months. Each is significant. Together they are a phase change.
The first is that building software got dramatically cheaper. Not slightly. Roughly a thousand times cheaper, depending on what you measure.
A working product that needed six engineers and a year in 2023 now needs one capable person and a few weeks. Tools that cost millions cost thousands. The whole stack to run a real business sits at a few hundred pounds a month.
This has not happened before. The personal computer revolution dropped the cost of doing computation. The cloud revolution dropped the cost of running infrastructure. Both took a decade or more to play out. The collapse in the cost of producing working software itself, which is what we are living through right now, is happening in months.
A note of honesty. The build cost has collapsed. The cost of running a real business has not collapsed by the same factor. Distribution, trust, customer support, regulatory compliance, all of these still cost what they always did. A solo operator can ship in two weeks what used to need a team. They still need to find customers, support them, and stay alive long enough to keep the thing running. The new economics do not erase the old work of business. They remove the build cost as the dominant constraint.
That is the build side.
The second shift is that the rails to charge for narrow value finally arrived. Boring sounding. Not boring.
The internet was never built for small payments. Card networks have minimum fees, usually around twenty or thirty pence. Bank transfers are slow. PayPal takes a chunk. If your product saves someone fifty pence, you cannot reasonably charge fifteen pence for it. The maths did not work.
That invisible constraint shaped everything. It is why subscription pricing became dominant. Aggregate lots of small uses into one big monthly charge that is worth processing. Anything genuinely small got eaten by transaction costs.
That problem has been solved, by the people you would expect.
The rails arrived
In May 2025, Coinbase launched a protocol called x402. It does something the internet has never done before. It lets software charge other software in fractions of a penny and settle in seconds. Not as a hack. As a native part of the stack. By May 2026 it had processed more than 169 million payments across over 590,000 buyers and 100,000 sellers in roughly twelve months.
In February 2026, Stripe added support for it.
Google launched a parallel protocol, the Agent Payments Protocol, with over a hundred partners.
In April 2026, Visa joined the x402 Foundation when the protocol moved to the Linux Foundation, alongside Mastercard, Microsoft, Shopify, and a coalition of more than twenty other industry participants. Visa, the largest payment network in the world, also launched its own command line tool for agent commerce in March, and its stablecoin settlement pilot now runs at a seven billion dollar annualised volume. The companies whose business depends on the existing payment infrastructure are publicly aligning themselves with the rails that bypass it.
Three days ago, AWS launched payment infrastructure for autonomous agents in partnership with Coinbase and Stripe. Companies including Warner Bros. Discovery and Thomson Reuters are already using it.
Stablecoin market value sits at around three hundred and eighteen billion dollars. The president of the European Central Bank has started spending speeches worrying about it. That is what financial infrastructure looks like when it stops being a hobby.
Read that list of names again. Coinbase. Stripe. Visa. Google. Amazon. The companies that built the previous internet are racing to build the next one. Not because they want to disrupt themselves. Because they can see what is coming and they would rather own the next layer than watch someone else own it.
To be clear about what this is and is not. The rails are real. They work. They are also early. 169 million transactions is a meaningful signal but a small fraction of what global commerce will need them to handle. Some of the protocols listed here will not become standards. Some of the names will not be the names that win. There will be a messy settling out period before the rails feel as obvious as card networks do today. The direction is settled. The detail is not. And the new substrate creates new attack surfaces that nobody has fully mapped yet, which is its own conversation for another piece.
Worth noting too that the same companies building the rails will extract rent from them. The infrastructure providers that defined the previous era are positioning to charge tolls on the next one. The new operators will earn on a much larger surface than was previously available, but they will pay for the privilege. This is honest rather than alarming. Every economic transition has its toll booths. The rails are still a structural improvement on what came before.
There is a cultural lag too. Technical feasibility is not consumer adoption. The energy bill scene at the start of this piece is technically possible now. It is not yet how most people pay their bills, because trust in autonomous agents takes longer to build than the agents themselves. The infrastructure is moving faster than the human comfort with it. That gap will close. It has not closed yet.
But the direction is the thing. The plumbing of the commercial internet is being rebuilt right now, in coordinated motion, by every company that defined the previous era. This is the substrate that turned your energy bill scene from impossible to inevitable.
The plumbing changed. Most people have not noticed. If your daily life still looks like it did two years ago, that is not a sign nothing has changed. Consumer experience usually lags infrastructure by twelve to twenty four months. The rails only became real in the last six. What you do not yet see is being built underneath.
A new kind of utility
Here is the way to understand what these two shifts produce when they happen together.
Software used to be sold like a product. A licence. A subscription. You paid for access, then you used it however much or however little. The price was set in advance. The relationship between what you paid and what you got was loose at best.
Intelligence is going to work more like electricity.
You do not buy a generator. You do not pay a flat fee whether you use power or not. You plug in. The power flows. The meter ticks up by exactly the amount you used.
Apply that to software. The agent that sorted your energy bill did not visit a website. It called a usage analysis function for a tenth of a penny. It called a tariff check for two pence. It read the small print for half a penny. It triggered the switch for twenty pence. Each of those is a piece of software somewhere, run by a different company, charging a tiny amount, settling instantly. No subscription. No login. No contract. Just the task, the cost, the result.
This is not a small change. It is the same kind of phase change that happened when electricity went from a curiosity to a utility, roughly a century ago.
Before the grid existed, every factory built its own generator. Most of life ran on steam, coal, human muscle. Then the grid arrived. Anyone could plug in. Within a generation, industries existed that nobody had imagined. Refrigeration, which transformed food. Radio, which transformed everything else. Domestic appliances. Modern manufacturing. None of it was about electricity itself. All of it was downstream of electricity becoming cheap, reliable, and pay as you go.
Intelligence is doing the same thing right now. Faster. The transition from curiosity to grid utility took electricity about thirty years. The equivalent transition for intelligence is happening in three to five.
This is what people miss when they call AI a productivity tool. AI is not a tool. AI is becoming a utility, in the same sense electricity is. Tools are bought, owned, used. Utilities are tapped, metered, drawn upon as needed. Tools shape their owners. Utilities shape entire economies.
We have spent two years asking what AI tools can do. The more useful question is what economies emerge when intelligence becomes a metered utility. We are about to find out.
What every company is already sitting on
Most executives have not seen the second order consequence of this. It is the most commercially significant fact in business right now.
If your business has been running for more than a few years, you are sitting on something the new economics suddenly value. Not the data exactly. The pattern inside it. Twenty years of customer behaviour. Decades of supplier relationships. Operational know how. The specific texture of how your industry actually works.
Until eighteen months ago, that pattern was stuck. You could use it internally. You could not realistically turn it into products. The build cost was too high. The markets you could reach with what you knew were too small. So the value sat there. Used by your own operations. Never deployed at scale.
The thousand fold drop changes that completely.
Pick almost any established business and the same thing applies. Twenty years of demand patterns. Thirty years of decision history. Decades of operational pattern recognition. None of it could previously be turned into products. All of it can be now. The data was always there. The capability is suddenly there.
This is the largest underpriced opportunity in business right now. It is invisible on most AI roadmaps because most AI roadmaps are framed around cost reduction. The conventional question being asked in boardrooms is “where can we use AI to do existing work more cheaply?” That is the small question. The structural question is different. What capabilities have we accumulated over decades that we could not previously turn into products, and which can now be packaged, priced, and deployed at scale?
Most companies will not ask this question soon enough. Not because the leaders are stupid. Because internal politics, legacy systems, sunk cost in existing roadmaps, and the ordinary risk aversion of established organisations make the path from “we have this data” to “we ship a new product line” genuinely hard. The companies that move first will be the exceptions, not the rule. Their advantage will be enormous and largely uncontested for a window of perhaps twenty four to thirty six months.
After that, the gap closes either through capable competitors who got there first, or through former employees who saw the opportunity their employer did not, or through new entrants with no legacy to defend. Companies that miss this transition will not slowly decline. They will be undercut at the edges and then in the middle by smaller competitors operating on the new economics. This has happened before. It happens fast.
What gets built that did not before
The new commercial surface is much larger than the one we have been living inside. Larger by something like four orders of magnitude when you multiply the factors together.
What fits inside it? Almost everything that did not fit before.
A coordination tool for the volunteer fire service in rural Cumbria. An app that handles the specific paperwork of getting a deceased relative’s affairs in order, which currently takes the average family between forty and one hundred hours of unpaid labour. Translation tools for the languages of the British deaf community, which sit below the threshold of every commercial translation product. A medication reconciliation agent for elderly patients leaving hospital, which is currently a leading cause of preventable readmission. A planning permission interpreter for self builders. Decision support for the parent who has been told their child needs an EHCP and has no idea what one is.
None of these are unbuildable. They were uneconomic. The cost of building was higher than the markets could support. Both constraints have lifted.
Dario Amodei talks, when he talks about why this matters, about the consumer surplus that arrives when access becomes broad. He uses cures for diseases as the headline example. The same logic applies all the way down. Every problem that nobody has had the resources to fix, that affects too few people to attract the previous era’s capital, is now within reach. Most of it does not look like cures for cancer. Most of it looks like the boring, specific frustrations that absorb hours of every life.
Both ends of the spectrum have opened.
The question to start with
If the piece has done its work so far, you are probably asking what any of this means for you specifically. Two questions are worth sitting with. They are not a checklist. They are a different way of looking at what you already have.
If you run a business of any size, the question is this. What does your business know that nobody outside it knows, and what would it take to package that knowledge into something other businesses or other people could pay for? Not as a service delivered by your people. As a product that runs on the rails described here, prices in proportion to the value it produces, and reaches markets you have never been able to economically reach.
Most leaders, when they sit with this question for the first time, find at least two or three answers within an afternoon. The hard part is not finding the answer. The hard part is taking the answer seriously enough to act on it before someone else does.
If you are a person rather than a business, the question is different. What have you spent the most time inside? Not what you do for work in the abstract. The specific corner of the world you understand more deeply than almost anyone else, because you have lived in it for a decade or two or three. The thing that frustrates you about it that you wish someone would fix. The advice you would have given yourself ten years ago. The pattern you have seen play out a hundred times that nobody has bothered to systematise.
Whatever you came up with reading those questions, that is where the new economics give you leverage that the old ones did not. The leverage is real. Whether you take it is a separate question, and most people will not, for ordinary human reasons. The opportunity does not require everyone to act on it. It requires you to act on it.
Where the work goes
The dominant story about AI is subtraction. AI takes jobs. Humans become redundant. The future contains less.
The structural story runs the other way. AI eliminates a particular kind of execution work, and that displacement is real. It will hurt real people. Whole industries will be hollowed out before the new shape settles. Pretending otherwise is its own form of dishonesty, and the optimism in this piece does not depend on pretending. The displacement is genuine and the people caught by it deserve more than a paragraph of acknowledgement, which is more than most pieces about AI give them.
But at the same time, AI dramatically expands the surface of solvable problems by making it economically viable to solve things that were previously uneconomic. The expansion is much, much larger than the displacement, and almost none of it has been built.
The question is who builds it.
Not the youngest. Not the most technically savvy. Not the people most fluent at prompting. The people best placed to build inside this new surface are the people who have spent their lives accumulating deep knowledge of specific problems in specific corners of the world.
The retired GP who has spent forty years watching exactly what is broken in general practice. The teacher who has watched the same learning patterns develop in real children for two decades. The carer who knows which signs predict deterioration in elderly patients and which interventions help most. The translator who has spent a lifetime moving between two specific languages and understands the subtle texture that no general purpose system captures. The specialist mechanic who knows what nobody else does about a particular kind of vintage equipment.
All of them were carrying capital the previous era did not know how to value. Decades of pattern recognition. Domain specific judgement. Accumulated experience that could not be turned into anything scalable, because the build cost did not allow it.
It is suddenly deployable.
To be clear, most retired GPs will not become founders. Most experienced teachers will not ship products. The opportunity does not require everyone to take it. The structural fact is that the people who do take it now hold leverage they did not have before, and the maths now works in their favour rather than against them. That is new. It is also the part of the story most likely to be missed by the dominant AI narrative, because the dominant narrative is being written about Silicon Valley by people in Silicon Valley, and the people for whom this transition matters most are not in Silicon Valley.
The 67 percent rise in entrepreneurship after layoffs in the last reporting period was not a freak. It was the start of something. People with deep experience in something specific, given for the first time the means to turn that experience into products that earn. The leverage moved. It used to live in the ability to write the code or run the platform. It now lives in knowing what should be built. That is the kind of knowledge that takes decades to accumulate, inside specific lives, and most of it has been waiting for the rails that finally make it useful.
This is the part that should give you hope. Not because it is sentimental. Because the maths now works in favour of the people who actually understand the problems.
What we are moving toward
Pull all of this together and the shape becomes clearer.
For thirty years, the commercial logic of consumer technology was to capture attention and rent access. Build something free or cheap. Get scale. Monetise the scale through advertising or subscriptions. The downstream effect was an internet optimised for engagement and consumption. Most of what got built reflected that logic. Most of what did not get built was filtered out by it.
The commercial logic now emerging is different. Build something specific that solves a real problem. Charge in proportion to the value it produces. Settle in fractions of a penny if necessary. Serve markets of any size, including very small ones. Earn from outcomes rather than from access.
This is not a moral upgrade. It is not the technology becoming more virtuous. The companies are the same companies. The people are the same people. What changed is the unit economics. And when the unit economics change, what becomes worth doing changes too.
The old internet rewarded scale and capture. The new one rewards specificity and usefulness. The first built a world of platforms competing for your time. The second is going to build a world of solutions competing for your problems. Not perfect. Not utopian. Different in ways that, for the first time in a generation, point in a direction most people would call better.
We are moving from technology built to capture human attention to technology built to solve human problems. That is the shift. It is not sentimental. It is structural. It is happening because the economics finally allow it, and it would be happening with or without anyone’s permission.
To be clear, the old internet is not dying. Google will keep selling ads. Facebook will keep optimising for engagement. The platforms that built the previous era will continue to run on the logic that made them rich. The shift this piece describes is not a replacement. It is the emergence of a parallel internet running on different economics, alongside the existing one. The interesting question is which of the two captures most of the new economic activity over the next decade. The answer, on present trajectory, is the new one. The old platforms increasingly look like utilities. The growth is happening elsewhere.
Even brands and businesses that depend on the old logic are going to feel this. When agents do most of the comparison shopping, brand advertising aimed at humans gets less effective. Companies that have spent decades buying attention will need new ways to reach buyers, and most of those ways will look more like solving the buyer’s actual problem than capturing their gaze. The attention economy is not ending. It is being reshaped by the arrival of an alternative.
The question is not whether this transition occurs. The question is who participates, who builds, and who gets to decide what gets solved next.
Where the value gets made
There is a geographic implication here that almost nobody is talking about yet, and it is one of the most important.
The previous internet concentrated commercial power in five American platforms. That was not an accident. The advertising and licensing economics that drove the previous era rewarded scale, and scale rewarded whoever could capture global distribution first. Silicon Valley was structurally advantaged by the unit economics. Every other geography was structurally disadvantaged by them. For two decades, “compete with American platform scale” has been the answer to every question about why a European or Asian or African technology ecosystem could not produce the equivalent of Google or Facebook or Amazon.
The new economics work the other way.
When the moat is local context rather than global scale, the advantage flips. A solo operator in Glasgow, in Lagos, in Manila, in Berlin, who has spent twenty years inside a specific problem in a specific market, now has the same rails as a venture backed startup in San Francisco. Cheaper rails, in many cases, because the operator is not paying for fifty engineers and a marketing team. The advantage in the new economics is the kind of local understanding that the previous era systematically discounted, and the previous era’s geography of capital is irrelevant to it.
This matters in a way that goes deeper than national pride or economic policy. Silicon Valley cannot know what it is like to navigate the planning permission system in a market town in the West Midlands. It cannot know what specific frustrations a Welsh farmer faces with the new agricultural payment scheme. It cannot know which gap in the Italian healthcare system is driving elderly patients to fall through the cracks, or what the actual coordination problem looks like for a community midwife in rural Kerala, or what is broken about how small construction firms in São Paulo quote for jobs. It does not know these things because nobody from Silicon Valley has lived inside them.
The people who have lived inside them now have intelligence infrastructure available to them that, until eighteen months ago, only Silicon Valley could afford. The asymmetry has not narrowed. It has reversed. Local context is the moat. Local context is what nobody can replicate from a distance. The thousand fold drop in build cost combined with payment rails that work across borders means that, for the first time in a generation, the structural advantages of being close to the problem outweigh the structural advantages of being close to the capital.
For Britain, for Europe, for every geography that has been told for two decades that it cannot compete with American platform scale, this is the first shift in a generation that points the other way. The companies and operators who understand this earliest will build the next decade of valuable software companies in places the previous era treated as peripheral. Most of them will not be in California. Many of them will be in places nobody is currently watching.
The world this builds
It is worth pausing here to imagine what this produces, because the structural argument is easy to admire and harder to feel.
Picture the world a decade from now if this transition goes the way the rails and the economics suggest it will.
A scheduling system that finally works around the impossible hours of community midwives. A pricing tool that lets the small builder in your village compete on the strength of knowing his trade rather than being undercut by faceless platforms. A diagnostic assistant for the vet treating exotic birds in a corner of practice the major firms have ignored. The dyslexic eight year old in your local primary school using software designed by a teacher who watched the same patterns develop in real children for twenty years. The Welsh farmer with an agricultural payment scheme interpreter built by someone who has worked the same land. The elderly patient whose carer has a deterioration alert built by another carer who saw too many hospital readmissions and decided to fix it.
Most of this will not be glamorous. It will not feature on conference keynotes. It will not generate ten figure exits. It will mean that hundreds of small frustrations that absorb hours of every life, that have sat unsolved for decades because the unit economics did not allow them to be solved, gradually get fixed by people who lived inside them.
The aggregate effect of that is enormous. The world becomes materially less frustrating in ways that are too small individually to notice and large in total. The people who fix these things become recognisable rather than remote. They are people in your town, in your industry, in your community, rather than a class of platform billionaires you have never met. The relationship between people who have problems and people who can solve them gets shorter and more direct than it has been in a generation.
Daniela Amodei made the point recently that AI adoption currently skews toward college educated men in higher income countries, but it is the Global South that is most optimistic about what AI can do. That asymmetry is meaningful. The geographies that have been told they are behind are the ones most likely to leapfrog the previous era’s assumptions about what software is for and who it is meant to serve.
This is not utopia. Plenty of problems will remain. Some will get worse. The transition itself will be hard for the people displaced by it. But the world that emerges on the other side of this shift, if the structural argument plays out, is one most people would recognise as straightforwardly better than the one we have spent twenty years building.
That is what is worth walking toward.
What opens
Build cost has collapsed. The rails work. Every meaningful business is sitting on capabilities it has not yet seen. The long tail of human problems is finally addressable. Decades of accumulated expertise inside individual lives is suddenly economically deployable. The maths now favours the people who understand the problems they are solving.
Nobody had to become more virtuous for any of this. The unit economics moved. The internet of attention is being joined by an internet of solutions, not because anyone designed it that way, but because the costs of building and the costs of charging both finally dropped to the point where solutions became the more rational thing to build.
None of this is a prediction. Nobody can predict the future, and the people who claim to are usually the ones least worth listening to. What this piece describes is the structural shape of a transition that is already underway. How it plays out in detail will be shaped by choices, accidents, regulation, competition, and a hundred other factors nobody can model in advance. The optimism here is not a forecast. It is a description of what has opened up, and an argument that the people who move into it first are likely to find the maths working in their favour.
Three orders of magnitude is a number. It is also a description of how much has shifted in eighteen months. Most of what comes next is still unbuilt. The companies and the people who move first will define what the next decade looks like, and what kinds of problems get solved in it.
The opportunity is open. The economics favour the people who lived inside the problems. If that is you, the rest is a question of whether you act before someone else does.
If this resonated, subscribe. The next piece looks at how to find the specific capabilities your business is already sitting on, and the practical questions worth asking before someone else asks them about you.
Craig Hepburn is an AI strategist and Perplexity Fellow. Twenty years building at the frontier of digital, from Microsoft and Nokia to Art Basel and UEFA. Now building at the frontier of agentic intelligence.
Great insights, Craig. Enjoyed the read.