When The Road Owns The Destination

I asked Google something the other day and got the answer.

I cannot remember what I asked.

I also cannot remember who actually produced the information that made the answer possible.

That second part feels more important.

For most of the web’s life, discovery involved movement. You searched for something, found a promising result, clicked it, and arrived somewhere else - someone else’s website, someone else’s newspaper, someone else’s shop, someone else’s slightly deranged personal blog last redesigned in 2007.

The search engine owned the road. Someone else owned the destination.

I have been circling part of this change for a while. In *When Search Stops Sending: Designing for an AI-First Google*, I was interested in what happened when Google answered a question instead of sending somebody to the site that could answer it.

What I had not really followed through was what happens when that answer becomes somewhere you can stay.

Increasingly, the road does not lead you somewhere. It gives you what you came for.

And if what it gives you is not quite right, you can ask another question.

That part changes rather more than search.

I Started With Curation

The question I originally wanted to think about was much smaller.

With effectively infinite content, does curation become a viable business?

There is too much to read, watch, listen to and generally allow into your head. AI has made producing still more of it astonishingly cheap. So perhaps the scarce thing becomes judgement - someone deciding that this matters, that does not, read this, ignore those six.

I have written about this before, in *The Next Scarce Skill: Curation*. When production becomes abundant, the bottleneck moves towards selection.

There is just one fairly obvious problem with treating this as a prediction.

Curation is already an enormous business.

Google curates search. Amazon curates products. Spotify curates music. TikTok curates whatever TikTok has decided culture is doing this week.

The more interesting change is not that platforms curate.

It is that the curator can increasingly make the thing it recommends.

That is where the question starts getting awkward.

When The Roadmap Becomes The Destination

Search used to produce a list of possible destinations.

You chose one.

AI summaries collapse that journey.

Ask what happened in an election, whether a particular boiler is any good, why a company’s shares fell, or what critics thought of a film, and increasingly you can get enough of an answer without leaving.

Enough is doing quite a lot of work there.

The answer does not need to be as good as the original reporting. It does not need to be equally nuanced. It does not need to include every caveat, disagreement or strange little aside that made the original worth reading.

It just needs to satisfy the immediate need.

That is the difference between replacement and substitution.

Replacement implies equivalence.

Substitution merely requires:

“That’ll do.”

Pew Research found that when Google showed an AI summary, users clicked a traditional search result 8% of the time, compared with 15% when there was no summary. Links inside the summary itself were clicked in just 1% of visits.

If I no longer click, the substitute has worked.

But calling it a summary increasingly undersells what is happening.

A summary is static.

The new surface talks back.

If the answer is too broad, I can narrow it. If something is unclear, I can ask for another explanation. If I want the counterargument, the practical implication, a comparison with another idea, or a version tailored to my circumstances, I can stay exactly where I am and ask again.

The original article cannot do that.

The journalist cannot answer a hundred thousand readers individually.

The academic cannot sit beside everyone reading the paper.

The critic cannot explain the review differently to every person who disagrees with it.

The generative layer can approximate all three.

So the platform is doing something more interesting than intercepting the journey to the destination.

It is constructing a destination that rearranges itself around me while I am standing in it.

The first answer does not need to be perfect. It only needs to be good enough that I ask my next question there too.

And somewhere in that process something stranger happens.

I am no longer merely consuming a derivative of the creator’s work.

I am interacting with an approximation of access to the creator.

What did they mean here?

How does this relate to their earlier argument?

Why did they reject the alternative?

What would this mean in my circumstances?

The creator supplies the corpus.

The platform supplies the relationship.

The roadmap has not merely become the destination.

It has started pretending someone is home.

Amazon Essentials For Everything

Amazon gives us a useful physical version of this.

A marketplace has an extraordinary view of demand. It sees what people search for, what they compare, what they buy, what they return, which prices work and which features seem to matter.

Then, occasionally, it puts an Amazon Basics or Amazon Essentials version on the same shelf.

That version does not have to be the finest object in its category. The Amazon Essentials T-shirt does not need to be the culmination of several thousand years of textile development.

It needs to be decent, cheap enough, easy to find, easy to buy, and preferably delivered tomorrow.

The substitute does not need parity when the platform controls the shelf.

Digital goods make this considerably more interesting.

A shirt still needs cotton, machinery, people, warehousing and a lorry. Software is reproducible at something approaching zero marginal cost. Generative AI goes another step.

The platform does not even have to reproduce the product exactly.

It can reproduce the function.

It can ask what people actually wanted from the thing in front of them. A summary? A workflow? A recommendation? A planning tool? A piece of analysis? Access to the apparent expertise behind it?

Then it can build a functional substitute inside the environment the user is already in.

The product does not need to be copied.

What the product proved people wanted can be copied.

That puts pioneers in a slightly unfortunate position.

They discover the interesting thing. They take the risk. They fail several times. Eventually somebody finds demand.

At which point the owner of the surrounding infrastructure has learned something useful without having paid the cost of learning it.

This is not new. Retailers have been doing versions of it for decades.

AI simply makes the private-label version of a digital capability much cheaper to produce.

Amazon Essentials without the troublesome requirement for a factory.

The Baron And The Board

I keep calling the owners of these systems Barons, which makes Monopoly rather difficult to avoid.

The fit is better than it first appears.

Elizabeth Magie patented *The Landlord’s Game* in 1904. She was influenced by the economic ideas of Henry George, who argued that while people should own what they produced, the economic value arising from land should benefit society rather than simply accruing to whoever happened to own it.

Magie intended the game to make some of those ideas visible.

The history is wonderfully circular. The game spread informally through Georgist circles, universities and Quaker communities before Charles Darrow encountered a version and eventually sold his adaptation to Parker Brothers. Parker Brothers later bought Magie’s rights for $500. PBS’s *Ruthless: Monopoly’s Secret History* tells the whole rather extraordinary story.

The version that survived is, perhaps unsurprisingly, the one where you ruin your friends.

There is something almost too neat about using Monopoly here.

The game is not really about building the best little house.

It is about controlling enough of the board that everybody else’s ordinary movement becomes economically useful to you.

You do not need every square.

You need enough structural advantage that, sooner or later, people keep landing on you.

The modern Baron has an even better arrangement.

It increasingly owns the road people are travelling on, the map showing where they might go, and the ability to build its own version of the destination once it sees enough traffic heading there.

It does not need to make the best destination.

It needs to make one good enough that leaving feels unnecessary.

Another login, another subscription, another app, another paywall, another five minutes. None of these is individually terrible.

Together they are quite a moat.

Convenience does not have to defeat your concerns.

It only has to make acting on them inconvenient.

The Invisible Shelf

Amazon still shows you the shelf.

Its own product is sitting beside somebody else’s. You can compare them.

A generative interface can remove the shelf altogether.

You do not necessarily see the platform answer beside the Guardian, Reuters, a specialist publication and a researcher with an alarming amount of domain expertise and 413 followers.

You see:

Here is the answer.

There is no obvious moment at which comparison happens.

We tend to imagine markets as though everything appears beside everything else and quality eventually wins. But whoever controls discovery partly controls the comparison set.

The better thing can lose without ever being judged worse.

It may simply not be encountered.

The conversational layer makes this stronger still.

Previously, a weak answer might have sent you onwards. You clicked another result, opened another tab, tried a different source.

Now the easiest response to an incomplete answer is often not to leave.

It is to ask again.

Explain that differently.

Go deeper.

Compare the two.

What am I missing?

The platform can absorb dissatisfaction that once created traffic for somebody else.

Even imperfection can keep the user inside the surface.

The substitute does not have to answer everything.

It has to remain the easiest place to ask what comes next.

Too Efficient A Parasite

At this point the relationship starts to look parasitic.

Original work costs something.

Someone travels. Someone interviews. Someone researches. Someone experiments. Someone spends twenty years learning enough about an obscure subject to recognise when everybody else is talking nonsense.

The platform consumes the result, synthesises it, and serves enough of it to satisfy the audience.

The creator still bears the production cost.

The downstream layer captures increasingly more of the value.

Parasite seems like a reasonable word.

Except there is a problem.

A reasonably successful parasite should avoid killing its host. Ideally there are other hosts around if it does.

Here there is really only one.

The collective body of human cultural and informational production is a shared substrate. There is not a second internet around the corner, stocked with untouched reporting, research, criticism, music and increasingly strange blog posts.

If the platform extracts too efficiently, eventually the thing being extracted from becomes less healthy - less expensive reporting, less difficult research, less specialist criticism, less work whose commercial return depended on people actually encountering the person who made it.

This sounds like a fairly obvious limit on the system.

The parasite needs the host.

I was quite comfortable with that explanation for about ten minutes.

Then it stopped working.

When The Parasite Starts Reproducing

The problem is that synthetic content can replenish the substrate.

Not perfectly. Not necessarily in the same way.

But perhaps perfectly enough.

Existing culture becomes the input for more culture. Generation produces content, content produces exposure, exposure influences taste, taste influences demand, and demand feeds subsequent generation.

At that point parasite is no longer quite the right metaphor.

It starts looking more like an invasive species.

An invasive species does not merely feed on an existing organism. It reproduces inside the ecosystem, competes for the same resources, spreads quickly and can alter the environment in ways that make the environment increasingly hospitable to itself.

Synthetic content has some fairly impressive reproductive advantages.

It is cheap, fast, extremely variable and instantly distributable. It is also often produced by, or inside, the same systems that control exposure.

The native species is expected to compete with something that can reproduce a million times before lunch and has a favourable relationship with the park ranger.

This does not necessarily destroy the ecosystem.

More often it produces a different one.

That distinction matters.

AI Slop May Already Be Historical

This is why I am increasingly unconvinced by the amount of attention we give to AI slop.

There is plenty of it.

Terrible prose. Ghastly motivational images. Videos involving babies, Jesus, military uniforms and anatomical questions best left unanswered.

It is easy to point at this stuff and conclude that synthetic culture has a quality problem.

Perhaps.

But I suspect we may already be looking at a historical artefact.

“AI slop” describes a period when synthetic content is recognisably synthetic because the production method is still visible in the result.

Early CGI looked like CGI. Early websites looked like early websites. Early digital photography looked very digital.

Then the technology disappeared into normality.

Two things are happening to synthetic content at once.

The systems are getting better, and we are becoming accustomed to what they produce.

The boundary is moving from both sides.

So perhaps AI slop is not the future of synthetic culture.

Perhaps it is the daguerreotype phase - a slightly embarrassing thing we will later recognise instantly as belonging to the early years.

The more consequential future is not one where AI content remains bad.

It is one where it becomes entirely acceptable.

Good enough. Sometimes excellent. Cheap. Personalised. Convenient. Normal.

Bad synthetic content is annoying.

Good synthetic content changes the economics.

I Was Tempted To Retreat To Novelty

At this point I was tempted to retreat to novelty.

Perhaps this is the bit humans keep.

Except I am not sure why I believe that.

Human originality is built from exposure, memory, recombination, error and occasionally misunderstanding something in a productive way. We inherit forms from culture, bend them, combine things that were not previously touching, and sometimes produce something strange enough that somebody else notices.

Machines do not create in the same way.

But that does not make novelty obviously ours.

The evidence is already inconveniently mixed. Some studies of divergent-thinking tasks have found leading generative models producing more original responses than average human participants. Other comparisons still find humans ahead, particularly at the most creative end of the distribution.

None of that settles the philosophical question of creativity.

It does make novelty a rather unsafe place to build a moat.

And scale makes the question stranger.

A person can pursue ten ideas.

A generative system can pursue ten thousand. Or ten million.

Most can be rubbish.

That is fine.

Evolution is not particularly bothered by wasted prototypes either.

Generate variations, expose them, observe response, retain what works, mutate it, try again.

At that point novelty starts to look less like a protected human property and more like something that can participate in selection.

The Fitness Landscape

This might be a better way to think about the Baron’s power.

The platform does not need absolute control.

Structural advantage is enough.

Natural selection does not need a committee deciding which animal deserves another generation. Persistent selection pressure gets you there eventually.

Platforms create selection pressures too.

They can make some things easier to encounter, easier to consume, easier to share, cheaper, more familiar and better fitted to the surrounding environment.

Tiny advantages repeated enough times stop being tiny.

The Baron shapes the fitness landscape.

And unlike an actual landscape, this one can watch what happens and adapt itself.

Something succeeds. The system notices. Distribution changes. More variants appear. The successful thing becomes easier to encounter. Its characteristics become part of what the next generation is optimised against.

This does not require conspiracy.

Nobody at Spotify needs to decide that humanity will now prefer a particular chord progression.

Nobody at Google needs to issue a memo defining tasteful prose.

The system can simply optimise for engagement, retention, convenience, cost and whatever other metrics currently occupy a dashboard in California.

Some things thrive under those conditions.

Others do not.

The environment selects.

Of course there will be several Barons. Google, Amazon, Meta, Apple, OpenAI, and others we have not met yet.

That sounds reassuring until you remember the crabs.

I wrote about this in *The Crabs Keep Winning*. Evolution has an annoying habit of independently producing crab-like bodies because similar environmental pressures keep producing similar solutions.

Platforms are subject to similar pressures too. Reduce friction. Increase retention. Improve personalisation. Keep the user inside the ecosystem. Own more of the customer relationship. Lower cost. Increase predictability.

Several Barons do not necessarily give us several fundamentally different cultural systems.

They may give us several crabs.

Different shells.

Same general arrangement of legs.

We Liked It Because We Saw It

This is where the argument starts touching taste.

I have written about this too, in *AI Doesn’t Need Taste - It’s Defining It*.

Taste feels personal.

Sometimes aggressively so.

Tell somebody their musical taste is socially constructed and you will quickly discover how personally they take their social construction.

But much of taste is exposure and agreement.

We learn what looks right, what sounds sophisticated, what appears expensive, what feels dated, what counts as beautiful, and what everyone has collectively decided is absolutely ghastly until approximately six years later when it becomes interesting again.

Culture teaches us how to judge culture.

So if a platform controls a substantial part of the exposure environment, it acquires a structural advantage in shaping the agreement.

Not control.

That distinction matters.

The platform does not dictate taste.

It controls a meaningful part of the field in which value is negotiated.

And generative systems add something recommendation systems could not do quite so easily.

They can alter supply.

We see something because the system selected it. We like it partly because we saw it. The system produces more because we liked it. We become increasingly accustomed to what the system has learned to produce.

The measuring instrument changes alongside the thing being measured.

At that point even the question of whether synthetic culture has “degraded” becomes slippery.

Degraded according to whom?

If people consume it, enjoy it and increasingly form their expectations around it, the Baron may not particularly care that a cultural critic thinks the ecosystem has become nutritionally deficient.

It only needs to work.

What Do We Decide Is Valuable?

I do not know what remains scarce.

And I am increasingly suspicious of anyone who answers too quickly.

Production clearly does not.

Perhaps exceptional quality does, but competent quality is becoming abundant very quickly.

Novelty no longer feels like particularly safe ground either.

Significance might remain scarce, although synthetic things can acquire significance too. A fictional synthetic musician could have listeners, influence, history, controversy, imitators and eventually nostalgia.

At some point it would be rather pedantic to insist that none of that counted.

Perhaps provenance matters. Perhaps trust. Perhaps access. Perhaps status. Perhaps being able to say that a particular thing came from a particular person.

But even that is not valuable because of some immutable property hidden inside the work.

It is valuable because people agree that it is valuable.

Which is what value rather annoyingly tends to do.

Luxury persists remarkably well in a world full of cheap functional substitutes.

Mechanical watches remain expensive despite being objectively worse at knowing the time than the thing in your pocket. People buy hand-thrown cups even though the industrial mug has solved tea. Vinyl records have somehow survived the invention of being able to hear almost every piece of recorded music ever made while standing in a supermarket queue.

These things are not valuable because they maximise utility.

Origin matters. Process matters. Scarcity matters. History matters.

More accurately, we have socially agreed that these things matter.

Human-made content may acquire a similar status.

Not because human work is necessarily better.

That argument is far too fragile.

It might simply become valuable because human-made becomes a meaningful cultural category.

This came from someone.

That matters to us.

There does not need to be mass agreement. Luxury rarely requires everybody. It needs enough people willing to assign disproportionate value to a distinction.

At the lower end that distinction is often signalled rather loudly - a logo, a signature, a recognisable name. At the very high end the signal can become quieter because the people buying it already know what they are looking at.

The same could happen with creators.

The Semi-Celebrity Creator

This is where things get personally unpleasant for anyone who makes things.

If the artefact becomes easy to substitute, one rational response is to make the creator harder to substitute.

Not: this is a good essay.

But: she wrote it.

The writer becomes a recognisable mind. The academic becomes a public intellectual. The chef becomes a personality. The designer develops a following. The musician increasingly sells access to the world around the music.

The person becomes part of the provenance system.

A body of work matters because it establishes continuity.

This person thought this three years ago, then changed their mind, then became obsessed with something else, then wrote a peculiar article about supermarkets.

The history becomes part of the product.

And now the interactive substitute from earlier starts to matter again.

If the platform can give me a reasonably convincing approximation of access to that person, simply being recognisable may not be enough.

I can ask the model to explain their work to me. I can ask how one argument relates to another. I can ask it to apply their thinking to something they never wrote about.

For a large amount of everyday consumption, that synthetic access may be perfectly sufficient.

Which pushes actual access further upmarket.

The conversation. The event. The membership. The commissioned work. The workshop. The signed thing. The hour in the room.

The creator does not merely become premium because a human made the artefact.

Proximity to the human becomes part of the premium.

This is not necessarily celebrity in the old sense.

You do not need millions of people watching you get out of a car.

A few thousand people who specifically care that you made something might be economically more useful than millions who barely register that anybody made it at all.

That sounds almost pleasant until you notice the incentive underneath it.

Creators may increasingly survive not by making better things, but by making themselves harder to commodify.

That rewards some fairly particular behaviours - visibility, continuity, public identity, community, and a willingness to turn bits of yourself into the wrapper around the work.

Not everybody wants that job.

Some people rather liked making things.

Curation Comes Back Into View

Somewhere in all of this, curation comes back into view - although not quite in the form I started with.

Suppose human creators become a premium layer.

There are still too many of them.

I cannot meaningfully follow every interesting writer, researcher, musician, designer and mildly eccentric person who knows an alarming amount about seventeenth-century drainage.

So perhaps the object of curation changes.

The question stops being which five articles I should read and becomes which people deserve a place in my attention.

The curator is no longer simply selecting artefacts.

They are selecting authorities.

This person is usually early. That one is occasionally wrong but interestingly wrong. This one has become repetitive. That one appeared six months ago and is worth watching.

The machine can do much of the surrounding work. It can follow them, summarise them, surface what matters and provide the ersatz conversation around their work.

The curator decides whose voices deserve to be represented inside that environment.

That makes curation more consequential than filtering.

It starts conferring attention.

And attention is part of how cultural value gets made.

A respected curator repeatedly saying “this person matters” does not merely report significance.

It participates in creating it.

Which also means curation helps construct the premium layer itself.

The curator who repeatedly surfaces a particular writer, designer or researcher is not merely finding value. They are helping establish the social agreement that makes that person valuable.

Unfortunately, this is not an escape route.

The Barons curate too.

Google ranks. Amazon recommends. Spotify selects. TikTok amplifies.

And the signals produced by independent curation are visible to them.

A person starts attracting attention.

A scene forms.

A format works.

A particular style repeatedly survives selection.

The Baron does not have to originate the judgement to learn from it.

It can amplify it, incorporate it, reproduce aspects of it, or simply make the successful thing easier to encounter.

Curation returns, but as another contested layer inside the same system.

There are still other routes - friends, teachers, schools, clubs, scenes, a strange record someone puts on at two in the morning, a book sitting face-out in a small shop, someone wearing something you have never seen before.

Call it folk exposure.

Culture does not begin entirely inside platforms.

Thankfully.

But once something discovered through folk exposure needs to be resolved or amplified, we often return to the Baron.

Someone tells you about a band.

You search.

Someone mentions a writer.

You search.

Someone shows you a chair.

You search.

The spark can happen elsewhere.

The infrastructure frequently determines whether the spark travels.

If A Tree Falls Outside The Feed

People will continue making things regardless.

I am fairly sure of this.

Humans were making little animals out of bits of bone long before anybody discovered a viable subscription model.

Creation itself is probably safe.

But culture is not merely creation.

Culture requires circulation.

A song sitting unheard on a hard drive exists. An essay nobody reads exists. A film nobody sees exists.

But does it participate in culture?

Not really.

Not until something happens between the artefact and other people.

Someone encounters it, talks about it, copies it, hates it, loves it, misunderstands it or builds on it.

That is when the thing becomes socially real.

So the Baron does not need to own every act of creation.

That would be an unnecessarily ambitious project.

It needs structural influence over enough of the machinery through which creation becomes visible, discussed, significant and valuable.

You can make anything you like.

The more interesting question is whether it can become culturally consequential without eventually crossing one of the roads.

The Whole Loop

This is quite a long way from where I started.

I wanted to know whether curation might become a business.

What I have ended up wondering about is the loop through which culture acquires value.

A thing is created, distributed, discovered, curated, synthesised, personalised and consumed. The reaction becomes feedback. Feedback influences taste. Taste affects what gets made next.

Generative platforms can increasingly participate in every part of that loop.

They can make the thing, decide whether you see it, decide how you encounter it, observe what you do with it, learn what people like about it, produce thousands of variations and use the resulting behaviour to influence what gets made next.

They can also stand between you and the creator while giving you a sufficiently convincing approximation of access to them.

The creator makes the work.

The platform explains it, adapts it, discusses it with you and answers the next question.

The road, the roadmap and the destination increasingly belong to the same system.

And so does more and more of the apparent relationship with whoever built the destination in the first place.

None of this requires the Baron to control culture.

Structural advantage is enough.

Natural selection managed quite a lot without anybody being in charge.

None of it requires synthetic content to become objectively superior either.

A merely sufficient substitute remains an alarmingly powerful competitive strategy when you own the route through which people encounter alternatives.

And none of it requires human creation to disappear.

People will keep making things.

Some of those things may become premium precisely because a person made them. Some creators may become semi-celebrities whose identity is inseparable from the work. Some curators may become trusted selectors of those people. Actual access to those creators may itself become a luxury good once synthetic access is available to everyone else.

Perhaps entirely new prestige systems will form around provenance, direct access and human origin.

Perhaps not.

That is the bit I am least interested in pretending to know.

The much stranger possibility is that quality, novelty and cultural significance may not remain safely human categories at all.

Synthetic culture might become good.

It might become strange.

It might become meaningful.

It might become culturally important.

It might become valuable simply because enough people agree that it is valuable.

Which leaves a fairly dark question for anyone who makes things.

If creation itself becomes cheap enough to commodify, do creators survive by making better things?

Or by making themselves harder to commodify?

And behind that sits a larger one.

If the road can build the destination, decide which destinations you see, and increasingly speak on behalf of whoever built them -

what remains outside the loop?