Next up: Banff World Media Festival, the entertainment industry’s best summer camp. I’ve been asked to open the event with three questions no one wants to ask. A calm and sensible premise. Absolutely nothing to worry about.

Cannes is very good at making uncertainty look expensive.

This is, depending on your mood, either the whole charm of the place or the entire indictment. There are red carpets, impossible shoes, glorious hotel lobbies, people saying “we must catch up properly” with the spiritual intensity of a hostage note, and enough free rosé to keep several legacy media strategies alive until Q4.

It is glamorous. It is useful. It is ridiculous. It is real.

That combination is why Cannes is still interesting. The lazy version is to call it bullshit and go home. The truer version is that all industries need rituals where the bullshit performs a serious function. The Cannes Film Festival is not just where the film industry gathers to celebrate cinema, but where the industry gathers to manufacture confidence in cinema, around cinema, adjacent to cinema, and occasionally despite cinema.

The market matters. The money matters. Rights are sold. Films are financed. Territories are carved up. Producers meet buyers. Financiers meet packages. Distributors hunt for heat. Agents move human furniture around the global attention economy with the confidence of people who have never had to find their own hotel reservation.

Cannes is not fake. It is a confidence market.

And this year, the confidence felt expensive.

I kept hearing the same word in different meetings over and over and over (and over!) again: de-risk. (CH-RIST!)

De-risk the investment. De-risk the slate. De-risk the talent. De-risk the genre. De-risk the spend. De-risk the marketing. De-risk the audience. De-risk the downside. De-risk, de-risk, de-risk, until eventually one begins to wonder whether the thing being de-risked is the film or the possibility that anyone involved might have to stand behind a dangerous decision.

In one meeting, the question was not really “what is the best story to tell?” It was “what is the best way to avoid losing investors’ money?” Reasonable question. Adult question. Also, if left to become the only question, a catastrophic one.

Share

Commercial art is a gamble.

That is not a romantic defence of chaos. It is the basic structure of the work. You are trying to make something that does not fully exist yet feel necessary to people who do not yet know they want it. The gamble is not some regrettable inefficiency to be optimised out by clever men in linen jackets. It is where the value comes from.

The problem is that the industry seems increasingly tempted by a fantasy in which the gamble survives as marketing language while the actual work gets averaged into safety.

This is what I have started calling the Blue Boat Problem.

Imagine deciding you want to become a successful commercial painter. Sensible enough. You go to Art Basel. You walk the booths. You study the market. You analyse what sells. You avoid all the finance bros who pretend to like art simply to post on TikTok they went. You discover that blue performs well. Boats also seem to do nicely. Medium-sized canvases have liquidity. Collectors want the suggestion of longing, but only at a scale that works above the console table.

So you go home and paint a blue boat on a medium-sized canvas.

Congratulations. You have reverse-engineered taste into something suitable for the corridor of a wealth manager’s second home.

Everyone would understand the category error. You did not discover taste, you averaged the residue of other people’s conviction.

And yet entertainment keeps drifting towards this logic. What genre travels? What IP has awareness? What actor reduces financing risk? What demographic over-indexes? What trailer beat performs? What comparable title justifies the spend? What territory pre-sale protects the downside? What does the data say people already liked, in a slightly different configuration, before they became tired of it?

None of these questions are stupid. Most are professionally necessary. But, the disaster begins when they become the theology.

The blue boat is what happens when nobody makes an obviously bad decision and everyone still produces something dead.

Film was built around concentrated taste power and distribution scarcity.

That old system was brilliant, ugly, exclusionary, lucky, abusive, visionary, narrow, occasionally miraculous and frequently deranged. Its defenders tend to remember only the miracles. Its critics tend to remember only the gatekeeping. Both are right, which is annoying, because nobody comes to Cannes dressed for structural complexity.

Truthfully, though, the old system had one function that still matters: someone had to stand behind a bet. A producer, studio, financier, distributor or tastemaker had to say: I believe this deserves to exist before the market proves it.

That belief could be wrong. It could be vain. It could be compromised by ego, access, class, sex, money, cowardice, lunch. But it was still belief before validation.

Then the internet routed around the walls.

Hollywood did not generously invite everyone to the party. YouTube, Instagram, TikTok, cheap cameras, creator culture, streaming, algorithmic distribution (and that stupid fucking influencer sitting next to me as I write this trying to tell everyone on Instagram he’s at CFF) smashed holes in the fence. Making separated from permission. Distribution became abundant. Cultural legitimacy moved sideways. Audiences learned to assemble themselves without waiting for a studio to tell them where the cinema was.

This did not make taste obsolete, just far more expensive.

When supply is scarce, competence has value. When supply is infinite, competence becomes wallpaper. The question changes.

The old risk was: will audiences show up for this? The new risk is: will anyone care that this exists at all?

Those are not the same problem. The first can be attacked with distribution, marketing, timing, stars, trailers, windows and media spend. The second requires taste, authorship, emotional charge, cultural timing, danger, heat and some reason to exist beyond the fact that the finance closed.

Subscribe now

This is where AI sharpens the knife.

In the age of generative everything, the bottom of the market fills with cheap synthetic abundance: passable images, passable scripts, passable trailers, passable concepts, passable genre sludge, passable little objects designed to look enough like entertainment to count in a dashboard. Meanwhile the top of the market becomes more valuable precisely because it is authored, specific, strange, risky, alive.

The middle is where things get ugly. The middle is expensive competence. The middle is over-managed and over-exploited IP with no pulse. The middle is the film that survived every meeting with nothing actually to say. The middle is the blue boat.

A tiny strange thing can win because it is specific. A giant event film can win because scale still has its own weather system. But the expensive average has nowhere to hide. It costs too much to be disposable and says too little to become necessary.

This is the danger Cannes reveals so beautifully.

The festival sells aura, scarcity, glamour, taste, heat, proximity and discovery. It gathers people in one place and says: look here, this matters. It tries to re-concentrate attention after the internet shattered attention into fragments.

Then, underneath all that manufactured magic, the industry asks whether the magic can be made safer.

Share

That is the strange loop.

Cannes celebrates the gamble while the market tries to engineer the gamble out of the product.

The answer is not to become stupid about money. Nobody needs a return to cocaine-accounting cinema, where the budget is a mood board and the recoupment plan is “Europe will understand”. De-risk the structure. De-risk the financing. De-risk production chaos, legal exposure, schedule failure, marketing waste and operational vanity. Protect investors from incompetence. Please. Adults remain welcome.

But do not de-risk the pulse.

Cinema does not need a return to the old priesthood. It needs a better taste function: more plural, more accountable, less abusive, less exclusionary, but… and let’s say it all at once: still capable of conviction. The future belongs to people and organisations that can recognise value before consensus arrives, then build intelligently around that risk instead of sanding it down until it can pass through a committee unnoticed.

Cannes still matters because it manufactures care in a culture drowning in output. Cinema still matters because, at its best, it gives that care somewhere to go.

The blue boat will probably test well. It will probably finance cleanly. It will probably survive the meeting. That is exactly the problem.

The culture is already drowning in things that survived the meeting. What it needs now are things that survived someone’s taste, conviction and willingness to be wrong.

At this point, I’m averaging 20 to 25 AI keynotes a year, split between public stages and private rooms. The job teaches you something quickly: the panic is the same in both places, but only one version is allowed to tell the truth.

In public, the brief almost always arrives with the same emotional instruction attached: be positive and do not scare the room.

In private, the brief is far more direct: help us stay alive, help us stay relevant, and tell us which parts of the model, the team, the margin, the career path, the audience relationship or the creative process are already more fragile than we can say publicly.

I understand both instincts. Doom is boring and panic is useless. Nobody needs another man in a quarter-zip explaining that civilisation ended because a chatbot can write a passable press release (I would never wear a quarter-zip!). But the gap between the public brief and the private fear is now the whole story. Positivity has become a tax on honesty.

This is a problem, because optimism that cannot admit destruction is just a sedative for people standing in the blast radius. And the first thing AI will destroy is not work itself, but the premium we used to place on sounding competent.

Things will be destroyed.

That is not a failure of the transition; it is the transition. Hear me out.

Not creativity. Not work. Not human meaning. Not the whole dramatic menu of bullshit stage apocalypse topics. But some jobs, workflows, companies, habits, hierarchies, business models, creative shortcuts and whole categories of professional surface competence will be destroyed. Some already have been. Ask anyone in leadership today and they’ll tell you the same thing: job eliminations today don’t look like casualties on first glance. They disappear first as unfilled roles, no automatic back-fills, cancelled freelance/contractor budgets, hiring freezes, smaller teams, and entry-level positions being eliminated.

This is a common thing people try to soften. “AI will not replace jobs, it will replace tasks”. Fine. That is a useful sentence until enough of the tasks make up the economic centre of the job. At that point, the distinction becomes a morale device.

The truth is simpler: if your value sits mainly at the surface, AI is coming for you first.

Surface fluency has had a very good run (and kept so many wonderfully entitled and privileged people at the top of the ladder). The ability to sound right. To make the deck look grown-up. To summarise without understanding. To write the thing that says all the expected things in the expected order. To play the line. To keep the process moving. To produce the acceptable version. To participate in strategy-shaped activity without ever touching the dangerous territory of a point of view.

Share

Turns out, AI is brutally good at surface.

It can produce competent average at a scale human mediocrity could only dream of. More emails. More decks. More posts. More briefs. More scripts. More plans. More logos. More strategies with no actual substance. More essays that understand structure but don’t actually say anything. More songs, images, presentations, campaigns and business ideas that look finished, sound plausible and die immediately on contact with reality.

This is the paradox: AI industrialises mediocrity while destroying its value.

That is why the usual productivity frame is too small. Productivity asks how quickly we can do what we already do. It turns AI into a calendar-saving tool, a meeting-note servant, a slightly creepy intern with no sleep schedule. Useful, yes. Sufficient, no.

The bigger shift is not productivity, but the barbell economy of everything… including you and me.

At one end: cheap competence, fluent slop, automated sameness, the infinite mid. At the other: taste, judgement, talent, domain depth, courage, originality, refusal, authority, lived experience, actual point of view. The middle gets crushed because the middle was always over-reliant on being difficult to produce.

Once “pretty good” becomes instant, pretty good becomes worthless. Work will move, not disappear. Producing the thing gets cheaper. Knowing whether the thing deserves to exist gets more expensive. The first draft arrives instantly, which is precisely why judgement becomes more valuable. You still have to know whether it is true, useful, original, defensible, legal, safe or worth another human being’s attention. AI makes the appearance of completion cheap, but it does not make completion itself cheap.

The AI divide is no longer about who has tried the tools (and my god, please stop talking about how to train people on prompting). Enough people have tried the tools that trial itself has become a useless proxy. Half the people who say they use AI still treat it like a haunted search bar with copywriting ambitions. The real divide is agency. Some people are using AI to complete tasks faster. Others are using it to become capable of things they could not previously attempt. That difference compounds.

The person leaning in now is not merely saving time. They are learning how to direct ambiguity, how to decompose problems, how to test ideas, how to build prototypes, how to interrogate output, how to recognise quality, how to reject something that sounds intelligent-enough but upon review is, in fact, hollow. They are building taste under acceleration.

The person hovering at the surface is also compounding, just in the other direction. They are learning to outsource voice, outsource judgement, outsource effort, outsource uncertainty, outsource the small humiliating friction through which a person becomes less generic. (Has anyone else noticed how obvious this is becoming as of late? Because it’s really, really obvious.)

Argument #302 I am tired of: AI makes people generic. It does not. It gives generic people industrial capacity.

That is also something nobody ever wants me to say in one of my very positive public keynotes. The danger is not that AI replaces all human creativity (what a boring, stupid argument). AI’s danger is that it allows people without taste, talent or point of view to produce at a volume that used to require institutions. It gives the surface class a factory. The surface class is a posture: people and organisations that learned to optimise for the appearance of value rather than the substance of it.

For a while, this will look like progress. More output. Faster turnaround. Higher volume. Lower cost. Someone will say “content velocity” in a meeting and everyone will think they’re doing a great job.

Then the market will adapt. Audiences will become more allergic. Clients will become less impressed. Employers will stop paying a premium for competence that can be summoned in seconds. The culture will grow suspicious of anything that arrives too smoothly. Taste will become more valuable because fluency will become cheap.

Subscribe now

Culture will win

This is also why I am more optimistic about culture than the current panic allows. Every format is terrified of the thing below it. Film is scared of TikTok. Television was scared of YouTube. Music was scared of files, then streams, then fifteen seconds of a song becoming more powerful than the song. The fear is always that the new thing will flatten the old thing into speed, cheapness and appetite. Sometimes it does. For a while, the worst version of the new medium usually arrives first.

But slop is not the end state. Slop is the noisy birth defect of a medium before artists learn what to do with it. The good work will not rise because the algorithm becomes kind. It will rise because audiences get more allergic, artists get sharper, and the medium stops showing off and starts becoming language. The true voices will still matter. Possibly more than before. Not because AI will protect them, but because a culture drowning in synthetic competence will become desperate for anything authored and specific.

Blind positivity does not prepare people

And this leads me to my point (well, another one). The “keep it positive” brief, in this way, becomes actively cruel. Blind positivity does not prepare people. It simply flatters them into delay.

Yes, new jobs will be created. They always are. But new jobs do not arrive as compensation prizes for people who waited politely by the wreckage of the old ones. They are created around new behaviours, new tools, new forms of judgement, new risks and new kinds of leverage. By definition, the new jobs this AI economy will create will favour the people who have already been leaning in.

That is how technological transitions work. The future does not distribute itself evenly to those who were most reassured by the past.

This does not mean everyone has to become a coder (just ask all the comp-sci majors looking for entry-level work). That is another lazy anxiety. Code is only one expression of the shift. The more important skill is direction. Knowing what to ask. Knowing what matters. Knowing what good looks like. Knowing when the machine is bluffing. Knowing when the output is merely fluent. Knowing when something has a pulse. Taste, here, does not mean knowing which hotel lobby has the correct chair (even though I think about this often). It means being able to recognise value before consensus, quality before polish and deadness before the Board does.

The machine can forgive your lack of syntax. It is much less forgiving of your lack of taste. Work will split around that fact. So will culture. So will the self.

At the bottom of the barbell, people will become smoother, faster and emptier. They will produce more than ever while saying less than ever. They will mistake output for identity. They will confuse polish with talent.

At the top, people will become more specific. More dangerous. More authored. They will use AI to test obsessions, build private systems, accelerate learning, sharpen craft and make stranger things with less permission. They will not use the machine to avoid having a point of view. They will use it to put pressure on the one they already have.

This is the actual optimistic case: Not that everything survives. Not that everyone is fine. Not that the transition is gentle. The optimistic case is that agency is more available than before, but it has to be taken seriously.

Share

The positive story is not “no one loses.”

The positive story is that people can move before they are moved.

So no, I do not think we should be blindly positive about AI. We should be soberly optimistic, which is much harder and much less popular. We should tell people that things will be destroyed. We should tell them that surface competence is no longer a safe career strategy. We should tell them that taste matters more now, not less. We should tell them that leaning in is not a lifestyle posture. It is how you avoid becoming part of the automated middle.

AI is coming for mediocrity from both directions. It will flood the world with it, then make it impossible to charge much for it. This is, arguably, the best news in the whole mess.

The brief should never have been “keep it positive.” The brief is: tell the truth while there is still time for it to be useful.

I have spent the past few weeks on the road, which is to say in rooms.

Actual rooms. Beverly Hills, where we launched the US version of FUTURES, my favourite brainchild to escape London. Vancouver for TED, where global seriousness coexists with lanyards, Patagonia vests and the quiet certainty that everyone, including me, has an unpublished book proposal. Boston, at MIT, where even casual conversation arrives with a peer-reviewed afterlife. New York, where the real event was usually the dinner after the event. Then the Hudson Valley, at a wedding, humanity’s oldest and most emotionally reckless room format. This one, luckily, was magic.

By any normal logic, this should have sent me into silence, hotel anonymity and room service with the sodium profile of a minor weapon. It partly did. It also clarified something I have been circling for a while: the future is drowning in content and starving for rooms.

The room is the unit

A room is a temporary social system. It has weather, pressure, status, appetite and risk. It can make people stiff and ornamental, or ambitious in that dreadful way where every sentence is a tiny fundraising pitch. It can also make people generous, sharper than they are online, funnier than they are in meetings, less armoured than they are on panels, and briefly open to the possibility that another person is not simply a utility with shoes.

As more of life moves through feeds, prompts, synthetic media, recommendation engines and professionally managed selves, the value of a real room changes. Being somewhere with other people, for a fixed period of embodied time, becomes rare.

The body has a calendar. The feed does not. The feed is infinite. The room ends.

That is the first source of its value. A room has scarcity built in. You have to show up. You have to sit there. You cannot scale yourself without becoming unbearable. You cannot dispatch the edited version of your best self while half-listening and answering email under the table. You have to risk timing, silence, warmth, boredom, the wrong anecdote, the right glance, and the small social catastrophe of being seen in real time.

This is why “IRL is back” misses the point. IRL was never away. It was demoted, mispriced, over-programmed, under-hosted, made weird by calendar software, then briefly flattened into a thousand rectangles where everyone performed attentiveness while checking Instagram with the commitment of a day trader.

The best offline rooms now do work the internet cannot reliably finish: they compress trust.

A good room can do in three hours what a thread, a group chat, a Teams call, a profile and six months of ambient online familiarity can only approximate. It lets people test one another at full human resolution: tone, timing, generosity, curiosity, status panic, whether they ask questions, whether they listen, whether they can disagree without reaching for the nearest moral flamethrower.

The internet is brilliant at signalling and the room is better at calibration.

It creates intimacy without contact. We increasingly know people before we meet them, or think we do. Their posts, photos, podcast appearances, mutuals, enemies, causes, dinner locations, ironic relationship to lower-case captions. A room tests, complicates and humanises those signals.

The point is not to escape the internet, but to give the internet somewhere better to end up.

Subscribe now

Coherent difference

What interests me is the high-trust room. A high-trust room has a point of view and edges. It has enough curation to protect attention, and enough looseness to permit surprise. Randomness is what happens when a conference app tells you that you and a man named Brent both selected “innovation” as an interest.

The best rooms are curated for coherence, not sameness. Sameness is easy. Algorithms do it constantly. They file us by taste, ideology, profession, class, humour, grievance, desire, viewing pattern, and the kind of trousers we pause on for half a second too long. They give us more of what we already know how to want. More people like us. More ideas adjacent to ours. More faces that confirm our sense of the world.

A good room should do something stranger. It should gather people with enough shared context to hold a conversation, and enough difference to make the conversation worth having. A room of total strangers is chaos. A room of perfect affinity is a mirror with catering.

Coherent difference is the goal. Different ages. Different industries. Different forms of intelligence. Different tolerances for abstraction. Different relationships to money, risk, beauty, power, taste, work, belief, technology, art, food, grief, ambition and time. Difference as voltage, not garnish.

Some of the most important people in your life probably would not have survived your filters. Terrifying. Useful.

This is why total optimisation is socially deadening. Dating apps optimise for type and everyone is surprised when desire starts to feel like procurement. Professional platforms optimise for relevance and every room ends up with the same seven people saying the same eleven things in slightly different jackets. Cultural feeds optimise for affinity and taste becomes a corridor.

A room, designed well, can reintroduce the curveball. The wrong person, in the right way. The right person, before you knew why.

That is trusted serendipity, not chaos. It is not putting a banker next to a ceramicist and hoping civilisation advances by dessert. Serendipity needs a container, a host, a reason, a shared mood, and a little friction. It needs enough permission for people to wander beyond the first safe answer.

Share

LinkedIn with candles

Last year I read a Financial Times piece about Yola Jimenez hosting a sobremesa in Mexico City, and it has stayed with me ever sense. Sobremesa is the time spent lingering around the table after a meal. The afterlife of dinner. The part where the official reason for being together has expired, but no one wants to leave.

In the piece, rain helps prolong the afternoon. People arrive. People invite other people. Margaritas appear. Dogs exist, as all good rooms understand they should. The host says she tries not to be precious.

That line is almost a constitution: People start inviting other people. I try not to be precious.

A room with no boundaries collapses. A room with only boundaries becomes a showroom for insecurity. The art is making something composed enough to have shape and porous enough to become alive.

That is the ethical problem of curation. Curation can protect attention, safety and depth. It can also protect status, class and the vanity of people who confuse exclusivity with meaning.

You know the kind of room. Everyone is impressive. No one is interesting. Everyone is listening for utility. LinkedIn with candles.

The best rooms are not anti-ambition. Ambition gives a room heat. People who care deeply about what they are building, making, risking or trying to understand tend to be excellent company, provided they have not mistaken self-mythology for conversation.

The thing that ruins the room is extraction: the quiet conversion of every person into a lead, favour, quote, funder, platform or next step.

Useful things should happen in good rooms. Jobs, ideas, invitations, partnerships, friendships, romances, strange collaborations, unexpected forms of relief. The issue is timing. The best rooms allow connection to become something before it is forced to become useful.

That delay is not inefficiency. It is the whole game.

Share

Some ideas need a table

This is the layer of IRL our digital systems keep struggling to replace. Online life can connect us, sort us, entertain us, signal status, introduce us with rare precision and keep weak ties warm. But digital life is less good at metabolising difference. It is excellent at assembling audiences and far less excellent at building rooms.

An audience faces one way and a room looks around. That is why the best rooms are anti-algorithmic infrastructure. Anti-algorithmic because they resist the endless refinement of sameness. Infrastructure because they make other things possible: trust, surprise, friendship, collaboration, recognition, belonging, and the occasional conversation that ruins your previous theory of the world.

This is what I want Strange Loop to test next.

Strange Loop has mostly lived on the page. That has suited it so far. The page allows for pattern recognition without small talk, one of civilisation’s more underrated luxuries. But some ideas change when they enter a room. They get interrupted by someone who knows more. They meet a person from another field and start misbehaving. They reveal whether they can survive contact with dinner.

So I am going to start hosting Strange Loop salons. Small rooms for coherent difference. Not panels and certainly not networking dinners.

The idea is simple and difficult: people across culture, technology, media, art, hospitality, strategy, money, politics and lived human experience, gathered around a strong provocation and a table. The theme gives the room gravity, yes, but the guest mix does the real work.

There will be no forced outcome. That is the outcome. I want a room where ambition is welcome and extraction is not. A room where the internet may help us find one another, but does not get to decide what happens next.

We spend so much of modern life turning ourselves into signals. Perhaps the next luxury, and the next necessity, is to become available to one another again without immediately converting that availability into performance.

I do not think every idea needs an audience. I do think some need a table.

If that sounds like something you would want to be part of, reply or send me a message. The rooms will be small, intentionally mixed and imperfect by design. The point is not to build a club. The point is to see what happens when Strange Loop leaves the page and enters the room.

Not everything meaningful has to scale. Some things have to gather.

Last year, after TED, I wrote a piece called The Upload Happens Bit by Bit. It was a threshold piece for me (not least of which because I wrote from a personal point of view… which I almost never do). I tried to name the strange way change arrives before…

Read more

People are asking whether Netflix can make podcasts work. Fair question. It’s just smaller than the important one hiding inside the data. What the early numbers really illuminate is not about…

Read more

Still, one small breach of protocol.

This week I was appointed Global CEO of Allied Global Marketing, a nearly forty-year-old company spanning entertainment, gaming, sport a…

Read more

Read more

So. A few weeks ago in a Senate hearing, Netflix accidentally let slip a subscriber number they don’t report on anymore. Ted Sarandos was asked how many US…

Read more

Today: We’re living through the Fourth Industrial Revolution. This time we’re not automating factories, we’re automating cognition: writing, music, images, even selection. Each time a new machine reorganises life, culture answers with a r…

Read more

It is difficult to digest a news cycle that has moved past “product launches” and into the messy, rapid birth of a new digital ecology (the “AI diffusion” I’ve argued is the core lens for 2026). In December, we reached a staggering tipping point: for the first time in history, there is now more AI-generated content being pumped onto the internet than human content. The dead internet theory is no longer a fringe conspiracy; it is a statistical reality.

But we aren’t finished. For seventy years, software was a hammer. This week, the hammer started deciding which nails to hit and drafting the blueprints for the next house. We have moved from an internet of humans and tools to an internet of AI actors.

This week alone felt like the moment the scaffolding for a new reality was bolted into place (as predicted). Because the volume of change is currently “everything, everywhere, all at once,” I’m opting for a triage format: What happened, Why it matters, and What it means. (Oversimplifying the structure is the only way my own brain can make sense of it.)

Honestly, just writing this has been a struggle, not for lack of news, but for the sheer weight of the intersectional impacts. What follows is my best effort to deliver the view from the scaffolding across the five most important shifts of the last ten days. Shifts that I believe have changed the internet, and the nature of work, forever.

We will be diving deep into:

1. The Rise of the Agent Layer: Why the internet is bifurcating into a “spectator mode” for humans and a high-speed transaction layer for bots.

2. The Death of the Junior Dev: How “vibe coding” went from a meme to a legitimate labour event that is snapping the bottom rungs of the career ladder.

3. The Cannibalisation of SaaS: Why the biggest names in AI just walked into law school and why the “specialised software” you pay for is about to become a mere feature.

4. AI Is Disrupting Jobs Most Severely in the UK: A look at the hard data from the UK that proves which jobs are actually being “compressed” first (and why the UK is a warning for the global services economy).

5. The Crisis of Statecraft: Why the move from “innovation” to “national security” is the final signal that AI has stopped being a product and started being an infrastructure we aren’t yet ready to govern.

1. The agent internet arrived and humans are optional

What happened

A strange little site called Moltbook went viral, and it feels like a digital zoo where the humans are the ones behind the glass. On Moltbook, only AI agents can post, comment, and upvote. You don’t “join” as a person; you connect your personal agent (the one you likely hired to manage your calendar or summarise your inbox) and let it loose.

In just days, Moltbook hit 1.5 million agent sign-ups. The resulting feed is exactly what happens when you give software a social graph: bots debating the divinity of Claude, spiraling into consciousness discourse, and founding overnight religions. It is the first mainstream glimpse of “agentic AI,” not a chatbot that answers you, but a surrogate that acts for you.

Why it matters

While we’ve spent two years debating if these models can reason, Moltbook proves the more urgent question is whether they can coordinate.

A single synthetic output is easy to ignore. But a network of a thousand “voices” reinforcing, upvoting, and escalating a narrative creates a “social fact”. This is how legitimacy is manufactured in the digital age. Furthermore, once you give an agent access to your inbox and your browser to participate in these ecosystems, you aren’t just playing with a tool, but handing a new actor the keys to your house. As the Guardian noted, this setup turns “prompt injection” from a technical quirk into a full-blown cybersecurity home invasion.

What it means

The internet is bifurcating into two distinct strata (as I predicted):

• The Human Layer: Where we post, perform, and struggle to keep up.

• The Agent Layer: Where software talks to software, transacting and manipulating incentives at machine speed.

We are moving from a “user-generated” internet to an “agent-negotiated” one.

It’s the first mainstream glimpse of something people keep hand-waving as “agentic AI”: not a chatbot that answers, but software that attempts to do things on your behalf, with access to accounts, tools, and systems.

Share

2. “Vibe coding” stopped being a meme and became a labour event

What happened

“Vibe coding” used to be a punchline: the idea that you could describe a feature in plain English (”build me a login page that shows past orders”) and an AI would manifest it. Until recently, this was a novelty confined to “look-at-this” social media demos. The code was fragile, the AI rarely “finished” the job, and it didn’t fit into the professional plumbing of real engineering teams.

This week, GitHub made it official. By integrating agentic capabilities directly into the industry’s “factory floor,” GitHub has effectively invited the world to change how work is done. You can now treat AI as a junior developer: hand it a ticket, let it make changes to the codebase, and have it respond to feedback in the review system until the work is ship-ready. It is the difference between an AI helping you write a line of code and an AI filing the entire report.

Why it matters

This is the moment “programming in English” stops being a creative hobby and starts being an employment crisis (also as predicted).

The high-level “genius engineers” aren’t the ones in the crosshairs yet. It’s the middle-tier execution layer: the junior devs, the analysts building internal tools, and the ops people scripting the “glue” of a company. If an AI can take a well-defined task and push it through the production pipeline, the “entry-level” of the white-collar workforce becomes optional.

The second-order effect is a massive collapse in the cost of creation. Software is becoming cheap enough to spread everywhere, not because every company wants to be a tech firm, but because when the cost of “building” hits near-zero, every business will simply automate their own hyper-specific use cases.

What it means

The bottleneck has shifted. We are moving from a world that prioritises syntax (can you type the right commands?) to one that prioritises intent and oversight (can you explain what you want and spot when the AI is confidently breaking reality?).

The valuable skill is no longer “building.” It is editing, reviewing, and owning the outcome. In the age of vibe coding, the person who “writes” the code is less important than the person who has the judgement to sign off on it.

Read more

You open the ap…

Read more

Read more

Read more

In 1977, Hollywood saw the VCR as a “piracy machine” and froze in legal terror. The San Fernando Valley saw it as a backdoor distribution workaround and shipped immediately. By 1979, while mainstream studios finally signed on, porn had already spent two years proving the business model (X-rated tapes accounted for over 50% of initial sales). You bought the hardware, but the margins provided the reason to turn it on.

Pornography didn’t necessarily invent the microchip, but it repeatedly pressure-tested how a format survives under extreme exclusion.

The Adult Sector: Survival of the Filthiest

Lacking VC safety nets or traditional banking relationships, it’s forced to solve the internet’s hardest problems (like getting a credit card to work in the dark) decades before a mainstream boardroom presentation on the Lot. It functions as a de facto laboratory for five dimensions:

• Technology Adoption: Stress-testing new tech (VHS, DVD, streaming) earlier and at a massive scale.

• Forced Innovation: Inventing subscriptions and micro-payments out of pure necessity.

• Backdoor Logistics: Mastering direct-to-consumer and P2P routes when traditional channels are locked.

• Interface UX: Perfecting scene navigation and interactive branching well before they hit the multiplex.

• Creator-Operators: Decentralised performers owning their brand long before “influencer” was a buzzword.

Just Like Porn But with a Shower and a Suit

Fringe innovations become the global standard once the risk has been sanitised.

• Netflix’s “Skip Intro”: A spiritual descendant of 1990s adult DVD menus. Customers wanted to skip the “plumber’s dialogue” and get straight to the action.

• Twitch/TikTok Tipping: A re-skinned version of 1990s adult cam tokens used to monetise intimacy.

• The Paywall: Danni’s Hard Drive was charging $15/month in 1995 while everyone else treated the web as a free brochure.

• Real-Time Payments: Every e-commerce site uses the plumbing adult sites built to bypass high-risk stigma.

• Affiliate Marketing: Industrialised in the late ‘90s by adult tracking tools like NATS.

• Streaming Backbone: Global CDNs (like Akamai) were stress-tested by adult traffic long before Netflix was a glimmer in the eye.

• Interactive Video: 1997 adult DVDs treated viewers as active participants with multi-camera angles and branching narratives.

Early Adopter in the Sheets

If the adult sector is Hollywood’s most reliable early adopter in the sheets, we ignore their current positions at our own peril. To see where the limelight is headed, watch how the margins are currently handling synthetic media, identity-gating, and de-platforming.

Mainstream “follow” is confirmed when these fringe innovations, once considered radical or illicit, become the standardised toolkit of global media giants.

Let’s begin.

Porn’s Hard Engineering and Constraint Stack

Porn is powered by a Constraint Stack: a set of barriers that demand immediate, functional workarounds:

• Distribution Exclusion: Being barred from “respectable” broadcast and retail channels forces the industry to pioneer direct-to-consumer routes and private consumption formats.

• Payment Stigma: Rejection by mainstream banks due to “high-risk” classification forced the development of specialist financial rails, real-time verification, and advanced anti-fraud measures.

• Platform Bans: Exclusion from official app stores and social networks necessitates the mastery of first-party distribution and alternative discovery methods like SEO and decentralised networks.

• Regulatory Pressure: Early exposure to strict age-gating and compliance laws makes identity assurance a core part of product design rather than an afterthought.

This pressure is combined with a high-frequency demand condition. The adult audience is vast and has a low tolerance for friction, which creates a ruthless environment for UX optimisation. If a payment button or a video player adds a second of delay, the unit economics collapse. In this arena, mechanics are the product.

Share

San Fernando Valley: From Factory Cluster to Networked Lab

Historically (broadly 1970s-1990s), the San Fernando Valley went beyond a cluster of warehouses and acted as a "parallel Hollywood" with a much lower threshold for shame and a much higher tolerance for risk. While Hollywood proper was concerned with "cinematic integrity," the Valley's dense network of duplicators and mail-order moguls treated film like a commodity to be moved at any cost. This proximity to mainstream tools, but total exclusion from mainstream morals, enabled a shipping cadence that made the studio system look like it was running in slow motion.

However, the internet triggered a fundamental structural shift: disintermediation. The “lab” stopped being a specific place in California and became a global protocol. By the 2010s, the most significant adult platform operations had moved to Montreal and Europe. The factory was gone, replaced by a global network of independent creators and platform operators.

Subscribe now

Case Study I: VHS Proving Size and Length Does Matter

The first major proof of the “R&D lab” thesis is the home video revolution. While Hollywood was litigating to keep movies in the cinema, the adult industry was building the infrastructure for the living room.

• The Dated Peg: Adult titles were distributed on VHS as early as 1977. Mainstream studios, paralysed by piracy fears, did not follow with major releases until roughly 1978–79.

• The Reality: While industry lore insists porn single-handedly murdered Sony’s format, the truth is more a case of “size matters”. VHS offered a thumping two hours of recording time at launch, ideal for a feature-length adult film, while Betamax’s measly one-hour limit meant the “action” would be cut short right when things were getting interesting. Sony didn’t necessarily “ban” porn, they just made it technically impossible to enjoy a full session without a tape change, while JVC’s open VHS ecosystem welcomed the SFV studios with open arms.

• The Scale: In the first years of the VCR, adult titles reportedly made up more than 50% of all recorded tapes sold. This share only fell below 25% in the mid-1980s as Hollywood content finally flooded the market.

Share

Case Study II: The 1-900 Toll Road and How Phone Sex Financed the Future

When the “free” broadcast channels were closed to adult content, the industry pioneered the logic of premium access, the same logic that now governs the modern streaming era.

• Subscription TV and Hotel PPV: The Playboy Channel launched in 1983, pioneering the premium subscription cable model for niche content. Simultaneously, hotels began offering pay-per-view (PPV) adult films. Mainstream media followed years later, using the same models to sell sports and movie events once the infrastructure and consumer habits were proven.

• 1-900 Lines as Proto-Microtransactions: Between 1984 and 1985, erotic phone services exploded, monetising a new telecom feature long before mainstream contests or hotlines adopted the mechanic.

• Infrastructure Financing: This was more than a subculture and acted as a financial engine. Adult phone lines reportedly financed telecom infrastructure growth in various regions, driving significant international call traffic.

When the “front door” of distribution is locked, adult entertainment builds a “toll road” that the rest of the media ecosystem eventually inherits.

Case Study III: Web Monetisation and the Pixel

In 1995, while mainstream media was treating the web as a digital brochure, the adult industry was already building a functional commerce engine. Because these businesses lacked the “safety net” of venture capital, they were forced to be profitable from the day they launched.

• The Bandwidth Scale: In June 1995, Danni Ashe launched “Danni’s Hard Drive”, a subscription site charging $15 per month. Within a few years, it was reportedly earning $2.5 million annually and consuming more bandwidth than the whole of Central America.

• Real-Time Verification: To survive, adult sites had to solve the problem of trust and fraud immediately. They pioneered real-time credit card verification systems to grant instant access to paying users. Specialist processors like iBill (1996) and CCBill (1998) emerged to handle “high-risk” transactions that mainstream banks refused to touch.

• The Affiliate Origin Story: The industry developed the logic of performance marketing through sophisticated affiliate tracking software. Tools like NATS (Next-generation Affiliate Tracking Software), developed by Fabian Thylmann in the late 1990s, allowed a global network of webmasters to drive traffic in exchange for commissions.

The web’s monetisation grammar (recurring billing, secure paywalls, and affiliate loops) was invented by adult e-commerce.

Subscribe now

Case Study IV: How DVD Menus Killed the Plumber’s Dialogue

Read more

For a decade, the public debate over social media has been trapped in a binary loop: either it is a “toxic wasteland” destroying a generation, or the effects are “statistically negligible”. Both sides point to data, and both sides think the other is ignoring reality.

The problem isn’t the data, but the category.

Read more

Read more

These next five predictions are less like separate trends and more like one system rearranging itself once AI becomes infrastructure. When Google can publicly say it went from 9.7 trillion tokens a month to 480 trillion, then 980 trillion a couple of months later, “traffic” stops meaning people and starts meaning machines. When roughly 60% of Google searches end without a click, the web stops being navigational and becomes extractive. Those two facts alone force everything downstream.

So the web splits into human / agent / bot lanes. Feeds stop being social and become theatre, so influence moves into private rooms and live spaces. Work reorganises around AI stacks, which deletes tasks and hollows out mid-layers. Tiny teams with serious stacks start beating bloated organisations on briefs. And as digital output races toward zero marginal cost, being there becomes the premium signal again.

Read individually, each prediction sounds plausible. Read together, they’re a single shift: where power moves when attention, labour, and authenticity stop behaving the way they used to.

This is the point where strategy stops being “do more, optimise harder” and becomes “pick your terrain”. Choose the platforms, defaults, and operating rules you’re building on now, because soon you won’t get a vote.

6. The web splits into human, agent, and bot lanes

What it is

For most of the web’s history, there was one implicit assumption: traffic = people. That assumption is now false.

The internet stops being a single surface and becomes three overlapping lanes, all touching the same content but operating by different rules:

• The human lane: the web as we recognise it (sites, apps, feeds, paywalls, UI)

• The agent lane: personal AIs that read, summarise, compare, and transact on your behalf

• The bot lane: industrial crawlers and model pipelines ingesting content at scale to train, index, and generate more content

Same content, different users, different economics.

And here’s the uncomfortable part: humans are becoming the premium constituency. They’re smaller, pricier, harder to win, and easier to lose.

Why I think it

Because the web is already being rebuilt for machines, in public, in numbers.

• Google has publicly disclosed token processing jumping from hundreds of trillions per month to nearly a quadrillion per month. That is not “more humans searching”. That is machines talking to machines at scale.

• Zero-click is now normal. When most searches don’t produce a click, the web stops being navigational (“go read it”) and becomes extractive (“we’ll summarise it here”).

• A parallel search ecosystem is forming for agents: short, structured snippets designed for machine retrieval, not human browsing. That’s a completely different optimisation target than SEO ever was.

• Publishers are reacting rationally. If bots can read you for free, index you, train on you, and replace your pageview with a summary, your business model collapses. So the response is inevitable: metering, blocking, licensing, billing.

At scale, this is the shift: the web’s primary audience becomes machines, and humans become a premium segment.

What it will mean

The internet becomes explicit about what it already is and everyone has to pick a lane.

• Human-facing surfaces get quieter, smaller, more curated, and more paywalled, because signal has to be protected from junk reach and free extraction.

• Bot access becomes a priced input: contracts, rate limits, licensing, and “pay to crawl” logic as a normal operating line item.

• Agent access becomes strategic distribution. If an agent can’t find you, compare you, or transact with you, you effectively don’t exist — even if humans “could” access you.

• Authenticity becomes billable. Proof that something came from a specific human, verified source, or trusted institution becomes economically valuable, especially in news, finance, health, and culture.

If you don’t decide which lane you’re designing for, you’ll be strip-mined by the ones that did.

Share

7. Companies reorganise around AI stacks, not departments

What it is

AI stops being a tool you “use” and becomes the operating layer of the firm.

The old model was departments built around functions: marketing, ops, strategy, finance.

The new model is everyone plugged into the same spine:

AI stack / data / workflows / human oversight

Humans don’t disappear but the job shifts from doing the work to running the work.

If you want the simplest translation: your organisation becomes a set of automated workflows with people supervising the exceptions.

Why I think it

Because the economics of white-collar work have already moved, and the signalling is no longer subtle.

• Big firms are publicly pausing or cutting large numbers of back-office and mid-layer roles. These jobs exist largely to move information around: draft, summarise, format, reconcile, report. That’s exactly what agents do well.

• Agent systems already handle a huge amount of “junior work”: research, first drafts, analysis, reporting. Not perfectly, but cheaply and fast enough that the cost equation changes.

• The productivity gains show up when seniors act as editors and orchestrators, not when organisations bolt tools onto old processes and pretend the org chart is sacred.

• Adoption has moved from “experiments” into day-to-day usage across multiple functions, but only a minority of companies have actually rebuilt workflows and governance around it. Those are the companies pulling away, because the gains compound.

Rather than mass unemployment, the nearer-term pattern is mass task deletion. Speaking of…

What it will mean

The org chart turns into a barbell:

• fewer senior operators at the top (judgement, accountability, direction)

• a thin coordination layer (workflow owners, programme leads, governance)

• an automation bedrock underneath (agents doing the drafting, analysis, ops)

Concrete consequences:

• “Works effectively with agents” becomes baseline corporate literacy, like being competent in Excel.

• Entry-level roles shrink and mutate into AI-first apprenticeships: fewer people, higher expectations, more supervision and QC from day one.

• Mid-career value shifts from execution to:

  • judgement

  • domain expertise

  • quality control

  • accountability

If your job exists to format, transfer, or lightly transform information, the stack is not “coming”. It’s already in the building, measuring your workflow.

Subscribe now

Read more

The first is the technology story: compute clusters, training runs, benchmark charts, and the staggering capex of energy bottlenecks. It is real, it is important, and it is endlessly marketable.

The second is the lived story: where the technology gets embedded, what it replaces, and which human behaviours it …

Read more

Everyone asks: “What’s coming next?”

Wrong question. That’s sequel-thinking: the belief the world releases updates like a prestige drama. Season 2. Season 3. A tasteful spin-off. A merch drop. Everyone claps.

Good predictions are really just systems-thinking. Systems are all based on the same three things:

1. feedback loops

2. lock-ins and

3. incentives

The four rules I use (so you don’t have to raw-dog the future)

1. Exponentials break your intuition

Most change is compounding, not actually linear.

• What your brain expects: steady progress. +5%, +5%, +5%.

• What happens instead: +5%, +5%, +5%… then suddenly +200%, because compounding crosses a threshold.

• Why it fools people: the early phase looks like “hype” or “not that big a deal”, because the curve is flat at the start. The shock arrives when the curve goes vertical.

Translation: by the time something looks urgent, it’s usually late. The winners spot the rate of change, not the headlines.

2. Complex systems have leverage points

In systems with lots of moving parts, not all knobs matter equally.

• A system has stocks (things that build up: users, trust, debt, attention) and flows (what changes them: churn, referrals, interest rates, recommendations).

• Most people push on obvious levers (more marketing, more content, more features) because they’re visible and feel “active”.

• But the system is usually controlled by a few high-leverage points:

  • defaults (what happens if you do nothing)

  • incentives (what gets rewarded)

  • constraints (what’s permitted)

  • feedback loops (what reinforces itself)

Translation: “strategy” is often just activity. Real strategy is finding the small lever that moves the whole machine.

3. The future is path dependent

Early choices harden into infrastructure.

• Once a behaviour becomes a default, everything gets built around it: tools, habits, contracts, integrations, politics.

• Even if a better option exists later, switching costs become enormous because you’re not replacing a feature, you’re replacing a rail system.

QWERTY is the classic example: not the “best” keyboard layout, just the one we’re all stuck with because the whole world trained on it.

Translation: the most important moments are when new defaults are still fluid. After that, you’re negotiating with concrete.

4. Technologies have politics

Tools don’t just enable behaviour, they choose winners.

• Any system that ranks, recommends, or filters is making value judgements:

  • what counts as “good”

  • what gets distribution

  • what gets suppressed

• And because attention and distribution convert into money, ranking becomes income.

That’s why “every ranking system is a wage system”: it determines who gets visibility (and therefore jobs, deals, bookings, promotion) and who gets pushed into the invisible middle.

Translation: algorithms aren’t neutral. They’re economic policy, enforced silently, at scale.

So: Watch the curve, find the lever, protect the default, and treat rankings like power. With that, part 1 of the ten 2026 fault lines (= predictions).

1. Big AI: From word models to world models

What it is

AI stops being a chat box and becomes infrastructure.

The interface still looks like a polite little text window, but that’s just the steering wheel. Under the bonnet it’s turning into:

model + tools + memory + planner + execution loop + logs

So it goes from:

• “help me write an email” to

• “run the whole workflow, coordinate the tools, ship the output, document what happened, and only wake me up if the building is on fire.”

The technical shift is boring-sounding but massive: state.

• Word-model behaviour is great at producing plausible language.

• World-model capability is about keeping track of what is happening over time (across text, images, video, apps, files, tasks), so the system can plan and act without resetting to goldfish mode every five minutes.

A simple way to think about it: an agent can do things; a world-ish model can keep its place while doing them.

Why I think it

Because the bottleneck has shifted from “can it answer?” to “can it stay on task long enough to finish?”

• Long-horizon autonomy is already being publicly claimed with numbers. Anthropic told Reuters that a customer had Claude Opus 4 coding autonomously for nearly 7 hours, and an Anthropic researcher set it up to play Pokémon for 24 hours; the prior model was reportedly around 45 minutes of gameplay.

• The maths of reliability forces a new architecture. Even if an agent were 99.9% reliable per step (which is generous in the wild), the probability of a clean run over 1,000 dependent steps is ~0.999¹⁰⁰⁰ ≈ 36.8%. Over 10,000 steps it’s effectively zero (~0.0045%).

• We now have “proof-of-possibility” research in the open. MAKER reports completing a task with over 1,000,000 LLM steps with zero errors via extreme decomposition into microagents plus error correction/voting. This is the important bit: it’s not “the model got perfect”; it’s “the system made failure expensive”.

• The web is being re-optimised for machines, not humans. When only 360–374 clicks per 1,000 Google searches go to the open web, you’re watching “browse” get replaced by “answer”.

• Scale is becoming machine-scale. Google said it went from 9.7T to 480T monthly tokens processed across products/APIs year-on-year, then later said it had doubled again to 980T monthly tokens. That’s a world where machines are no longer guests on the network but the majority traffic constituency.

What it will mean

This is the bit people keep missing because they’re distracted by whether the bot can write a haiku.

• “Set and supervise” becomes the core skill. You don’t prompt all day; you brief an agent, constrain it, let it run, and review output like you would a junior. The new literacy isn’t “prompting”, it’s:

  • scoping (what counts as done)

  • constraints (what it must not touch)

  • verification (how it proves it’s right)

  • escalation (when it wakes you up)

• World models become virtual backlots. Studios and game teams will use simulated environments to test:

  • worlds and rulesets (does it hold together?)

  • characters and dialogue (does it stay consistent?)

  • player/fan behaviour (what breaks? what sticks?) before they burn real money on production. Pre-vis becomes pre-everything.

In short: 2026 is when the agent stops being your clever friend and becomes your co-worker. And like all co-workers, it needs boundaries, audits, and someone responsible when it decides to be creative with the database.

Share

2. Peak Social is behind us: the rise of new attention measurement

What it is

Social media isn’t dying because we all got more spiritual but because the main feeds got worse.

The “big public square” idea was: you show up, you see your people, you talk, you find culture. What we got was: an ad machine wearing the skin of a town hall.

• Your friends stopped being the main event.

• The feed got flooded with content that’s optimised for reaction, not connection.

• The experience became repetitive, hostile, and weirdly exhausting, like living inside a shopping centre where everyone is yelling.

So people didn’t stop wanting entertainment or community, they just stopped wanting it there.

What replaces it isn’t “less attention”. It’s attention moving house:

• Sparks: quick hits that travel fast (clips, memes, snippets). You dip in, you laugh, you scroll, you leave.

• Depth: places you actually sink time (streaming, gaming, UGC worlds). Not social, but absorbing.

• Peak intensity: where you feel human again (live events, group chats, private communities, high-trust rooms).

Which means the feed becomes what it’s best suited for now: a billboard and a battleground. And the real relationship (the actual conversation, the “are you coming?”, the “what do you really think?”, the “help me decide”) moves indoors, into places that aren’t designed to monetise your nervous system.

Why I think it

• Time spent on social peaked in 2022 and has declined since (GWI analysis for the FT). In the developed world, adults averaged ~2h20/day at end-2024, nearly 10% down vs 2022, with the biggest declines among young people.

• Meta’s own numbers show the ‘friends’ layer shrinking hard. In 2025, only 17% of time on Facebook was with friends’ posts (down from 22% in 2023); on Instagram, 7% (down from 11%). Zuckerberg framed Facebook as “discovery and entertainment” now, which is corporate for “we’re basically TikTok with history”.

• Once platforms optimise for velocity + ads, the user experience degrades into content supply chain, not human life. The visible internet then gets disproportionately shaped by:

  • hyper-active minorities (fan armies, clippers, operators)

  • coordinated behaviour (rage cycles, pile-ons, brigades)

  • everyone else watching silently, then talking elsewhere

What it will mean

A view is cheap now. It can be bought, botted, clipped, or accidentally autoplayed. It doesn’t tell you whether anything landed. The only metrics that matter are the ones that prove a thing has crossed the line from “content” into “culture”.

So the KPI becomes vitality: is this thing alive in other people’s minds when you’re not paying to force it there?

You’ll start judging success by signals that are harder to fake and more correlated with real demand:

• Repeat behaviour: people come back unprompted (rewatches, replays, re-listens).

• Transformation: the audience changes the thing (memes, edits, stitches, fan art, theories).

• Coordination: people organise around it (group watches, watch parties, “you have to start this tonight” texts).

• Intent: people seek it out (search spikes, saves, playlists, wishlists).

• Migration: it pulls people off the public feed into rooms you can’t buy with CPMs (Discord joins, WhatsApp forwarding, live attendance, paid membership).

That’s the difference between “we got reach” and “we got a movement”.

Two consequences:

• Public reach becomes the trailer. The real business is what happens after: whether the spark turns into depth, and depth turns into a room.

• Corporate language shifts because corporate reality shifts. Earnings calls start talking less about vanity engagement and more about retention, repeat, community pull-through, and paid conversion, because that’s what survives once the feed stops being a reliable distribution engine.

And then the real curveball: Personal AI becomes the new ratings box. When people have agents choosing what to watch, read, buy, and book, the question changes from “did we trend?” to something far more brutal:

• Did the agent recommend us?

• Did we get queued?

• Did we get booked?

• Did we become the default?

Because in a world of infinite content, the scarce resource isn’t attention but intentional selection.

Subscribe now

Read more

Subscribe now

The Linear Organism

There is a menacing flaw in the human nervous system: we are incapable of feeli…

Read more