Sixty years ago, two groups from one family left the Portuguese island of Madeira looking for a different future. Six went to Brazil. Five went to Australia. Same island, same family, but two doors that would shape the lives of generations that followed.

Those families couldn’t migrate to Brazil AND Australia. They chose. They then lived the consequences of that choice for generations.

Why it matters

AI gives us more possibilities than any generation has ever had. But the ancient human requirement remains. At some point you still have to choose a shore.

I keep thinking about that story because possibility has always moved human beings. It has pulled us across oceans and borders, into new businesses, relationships, careers and adventures. Without possibility, we stay where we are.

And being free to choose a life a of freedom and seeking abundance sits inside all of us. Some of us act and others don’t choose or change.

But possibility only changes a life when it becomes commitment.

Today, AI is giving us more possibilities than any generation has ever had: more information, tools and ideas; more businesses we could start, careers we could pursue, skills we could learn and versions of ourselves we could become.

That sounds like freedom, and much of it is.

But abundance creates a new problem: when the doors keep multiplying, how do we decide which one deserves our finite attention? At some point, we still have to choose a shore.

From scarcity to abundance

For most of human history, the challenge was finding an open door. Information was scarce, expertise was expensive and opportunity was constrained by geography, money and access.

AI reverses that equation. It can suggest a business, research the market, write code, design a product, draft a book and generate ten alternative directions before we have acted on the first one.

That is extraordinary leverage. It also creates a new kind of overwhelm. The question is shifting from “Where can I find opportunity?” to “Out of everything now possible, what is worth choosing?”

Subscribe now

The Numbers

The numbers show how quickly possibility is scaling. AI is spreading through everyday work and life while dramatically lowering the cost of turning an idea into something real.

Figure 1. Source: Stanford AI Index 2026 / McKinsey organisational adoption survey

Stanford’s 2026 AI Index reports that generative AI reached about 53% population-level adoption within three years. In organisations, reported AI use in at least one function rose from 55% in 2023 to 88% in 2025; regular generative-AI use rose from 33% to 79% over the same period.

Figure 2. Source: Sensor Tower, State of AI 2026

Sensor Tower projects 36 billion hours spent in generative-AI apps in the first half of 2026, up from 17.2 billion a year earlier. More than 200,000 apps now mention AI and are on track for nearly 10 billion downloads in H1 2026. ChatGPT reached one billion monthly active mobile users in May 2026.

On the supply side, Dealroom tracks roughly 18,500 AI startups — before counting the millions of experiments, side projects and tiny businesses that may never seek venture funding.

Figure 3. Source: Lovable, “A first look at the build economy” (June 2026)

The deeper change may be who can build. Lovable says more than 60 million projects have been created on its platform since November 2024, and 80% of builders in its 2026 research identified as being in non-technical roles. The distance between “I have an idea” and “I can test it” is shrinking fast.

Figure 4. Source: Azoulay, Jones, Kim & Miranda, “Age and High-Growth Entrepreneurship”

Age is not the barrier many assume. A U.S. Census-linked study of 2.7 million founders found a mean founder age of 41.9, rising to 45.0 among the fastest-growing one-in-a-thousand ventures. Conditional on starting a firm, a 50-year-old founder was 1.8 times as likely as a 30-year-old to achieve upper-tail growth.

“For most of history, possibility was scarce. AI is making the cost of generating another possibility approach zero. Human time is not”.

The Abundance Paradox

More possibilities should make us freer. Up to a point, it does. But every plausible option asks for attention, every open door creates comparison, and every new tool raises the suspicion that the one we are using is already obsolete.

The paradox is simple:

AI can expand the possibility space almost without limit while the human using it remains finite. We still have twenty-four hours in a day, limited energy and attention, and only one life to allocate.

Herbert Simon saw the first half of this problem decades ago: information abundance creates attention scarcity. AI extends it. When intelligence and option-generation also become abundant, the scarce resources become judgment, focus and commitment.

More choice is not automatically bad. Research on choice overload finds that the problem depends on context. It becomes more likely when options are complex, the stakes are high, our preferences are uncertain and evaluation is costly. Twenty restaurants are one thing. Twenty plausible careers or businesses are another.

Research: Choice overload meta-analysis, Journal of Consumer Research

AI can also reduce effort while shifting cognitive work elsewhere. Microsoft researchers studying 319 knowledge workers and 936 real-world GenAI uses found that effort moved toward verification, integration and supervision; higher confidence in AI was associated with less reported critical thinking.

A 2026 study of 460 knowledge workers found cognitive overload and decision fatigue contributed significantly to intentions to stop using GenAI. The point is not that AI inevitably overwhelms us. It is that abundance without a way to filter, choose and stop can turn leverage into cognitive load.

Research: Microsoft Research on GenAI and critical thinking • 2026 study on GenAI, cognitive overload and decision fatigue

The new human problem is not discovery. It is direction.

The response to AI overwhelm cannot be only better prompts, workflows and tool stacks. Those sit downstream from a more important question: What deserves my finite attention?

We need a human operating system for moving from possibility to commitment: Direction → Focus → Persist → Renew.

1. Direction — decide what deserves exploration

Direction is not about discovering one perfect purpose. It means knowing enough about your energy, strengths, values, curiosity and circumstances to choose the next meaningful experiment.

Use AI to expand the map, but do not live in the map. Make the prototype. Interview the customer. Publish the article. Spend time with someone already living that path. Reality is a filter AI cannot replace.

2. Focus — choose a door

Once a direction has earned your attention, choose a door. The discomfort is real because choosing one path means temporarily refusing the others.

Keep a “Not Now” list. Capture new ideas without letting them interrupt the commitment in front of you. In the AI era, focus is not just concentration; it is the ability to exclude attractive possibilities.

3. Persist — stay beyond the novelty

AI makes starting extraordinarily easy. That may make staying unusually hard.

The first version is exciting. Then the work becomes repetitive, customers are difficult, the code breaks or the audience grows slowly. At precisely that moment, AI can produce another seductive path.

Before changing direction, ask: Has reality produced new evidence that this is the wrong path, or have I simply reached the unglamorous middle? Persistence gives compounding time to operate.

4. Renew — persistence is not imprisonment

Persistence is not imprisonment. Commitment needs review points.

After a defined period — seven days for a micro-experiment, thirty days for a project or a quarter for a strategic direction — ask what reality taught you. Did the work energise you? Did your skill grow? Did anyone care? Is there traction?

Then decide explicitly: recommit, adapt or leave. Do not drift.

“Direction chooses the horizon. Focus chooses the door. Persistence walks through it. Renewal decides whether to keep going”.

A routine for living with abundance

A framework matters only if it changes behaviour. A simple weekly rhythm can keep abundance from becoming distraction:

1. Start with the human question: What matters most this week before I ask AI what is possible?

2. Choose one primary door: give one project first claim on your best attention.

3. Use an option budget: ask AI for three serious alternatives, not twenty by default.

4. Keep a Not Now list: capture possibilities without turning them into commitments.

5. Set a stopping rule: decide what evidence is sufficient to act. More research is not automatically more wisdom.

6. Schedule contact with reality: every important direction needs an action that produces information from the world, not another AI conversation.

7. Review on a cadence: recommit, adapt or close the door consciously.

The Choose Your Door test

When several possibilities look attractive, use five questions. The point is not mathematical precision; it is forcing judgment into the process.

• Energy — Does doing the work give me energy, not merely imagining the outcome?

• Identity — Does this fit the person I am becoming and the strengths I want to use?

• Evidence — What signal from reality says this deserves another month of my life?

• Compounding — If I stay for three years, what compounds: skill, reputation, relationships, assets, knowledge or trust?

• Cost — What meaningful thing must I stop, postpone or decline to give this path a real chance?

Then choose. The goal is not certainty. It is enough conviction to move.

“Find your direction. Choose your door. Stay long enough to be changed by it”.

Subscribe now

Too many doors?

AI can show you more possibilities than you could ever pursue. Zyrro is being built to help you narrow them.

It starts with who you are — your patterns, strengths, energy and what matters to you — then helps you identify meaningful directions, choose one, and turn it into action.

Because in the age of AI, the problem is no longer finding opportunity.

It is knowing which door is yours.

Zyrro is coming soon. Join the waitlist to be among the first to try it.

Sources and further reading

European Commission regional portrait: Madeira — population and emigration

Regional Government of Madeira — Madeiran emigration to Australia

Arquivo Histórico da Madeira — emigration from Água de Pena, 1955–1970

NSW Migration Heritage Centre — Australia’s post-war migration, 1945–1965

Stanford HAI — 2026 AI Index, Economy

Sensor Tower — State of AI 2026

Dealroom — AI sector profile

Lovable — A first look at the build economy

Lovable — Series C / project scale

NBER — Age and High-Growth Entrepreneurship

Microsoft Research — Generative AI and critical thinking

Journal of Consumer Research — choice overload meta-analysis

Journal of Enterprise Information Management — GenAI, cognitive overload and decision fatigue

Cambridge University Press — attention scarcity and information abundance

Surrender of control does not arrive dramatically or is announced with trumpets. It arrives as a convenience.

A map saves us from remembering the route. A streaming service chooses the next song. A social feed decides which post deserves our attention. Search engines rank the answers before we see them.

Each choice is small. Each one feels sensible. Convenient.

That is why I have been asking myself a simple question: When does convenience become dependence?

I would now add a second one: When does dependence become delegated control?

Why it matters

That question matters because AI is changing the scale of handing over control to the machines. Delegation.

The first generation of digital tools mostly helped us find things, remember things and move faster. The new generation increasingly helps us interpret, recommend, decide and act.

There is nothing automatically wrong with that. Delegation is one of civilization’s oldest technologies.

We delegate more than we realize:

• Memory to writing,

• Arithmetic to calculators,

• Navigation to maps,

• Expertise to professionals

• Routine tasks to software.

The problem begins when we no longer notice what has been delegated.

We think we are still choosing because we approved the system once. But the system may now be shaping the menu, ranking the options, framing the problem and suggesting the next move without us even noticing.

At some point, the question changes from “Is this useful?” to “Who is authoring this decision?”

We have already delegated attention

Social media gives us an early warning.

A 2025 preregistered audit published in PNAS Nexus compared Twitter’s engagement-ranked feed with a reverse-chronological feed.

• The engagement system amplified more partisan content, more hostility toward political out-groups and more anger.

• For political tweets, anger expressed by authors increased by 0.75 standard deviations and anger felt by readers by 0.37 standard deviations.

• Partisanship and out-group animosity were each amplified by 0.24 standard deviations.

Figure 1. Effect sizes reported in the 2025 PNAS Nexus audit of Twitter’s engagement ranking.

The most important finding:

It was not simply that the algorithm amplified emotionally charged material. It was that “Users did not necessarily say they preferred the political content the engagement system selected for them”.

That distinction is enormous.

An algorithm can become very good at predicting what will make us react without becoming equally good at understanding what “we would choose” upon reflection.

Those are different objectives.

• Engagement is revealed behaviour: what I clicked, watched, shared or argued with.

• Preference is closer to reflective intent: what I say I actually want more of in my life.

A system optimised for the first can slowly overpower the second.

This is not just a social-media problem.

It is a design problem for the AI age.

If a platform can choose what enters our attention, an AI assistant can potentially influence what leaves it as a decision.

The delegation ladder

There is a useful way to think about how this happens. And it starts with the ease of convenience. Making it easy.

One way to put this in context is the chocolate bars placed right next to the checkout.

Convenient but dangerous.

And here is what that delegation looks like as a ladder.

The big question to consider is “At what point did I stop choosing?”

The danger is not that every useful technology inevitably climbs this ladder. Most do not. The important point is that each step can feel almost identical from the inside.

• The route recommendation is convenient until I no longer know the city.

• The feed is helpful until I stop choosing what deserves my attention.

• The writing assistant is useful until I cannot hear my own voice without it.

• The decision assistant is efficient until its default becomes my preference.

And because the movement is gradual, there is rarely a dramatic moment where we say: “Today I surrendered control.”

We simply stop exercising a capability because exercising it has become less convenient than accepting the recommendation.

This is why the most important AI question may not be “What can this system do?”

It may be “What do I want to remain able to do?”

That changes the design goal.

A good AI system should not merely minimise friction. It should help us decide which friction is waste and which friction is part of being a capable human.

We do not need to preserve every old skill for nostalgia’s sake.

But we should be very cautious about surrendering the skills that allow us to choose our direction, revise our beliefs and take responsibility for our lives.

From delegated attention to delegated action

The change sounds subtle, but it is profound.

• A recommendation algorithm says: “Here is what you should see next.”

• An AI assistant can say: “Here is what you should do next.”

That might be the next email to send, the candidate to interview, the investment to research, the article to read, the meeting to cancel, the business idea to pursue, or the way to respond to another human being.

Again, that can be enormously useful.

I use AI because it expands my field of view. It can surface possibilities I would not have considered, challenge an assumption, compress research and help me see patterns.

But abundance creates its own problem.

Ten plausible answers do not remove the need for judgment. They make judgment more important.

This is where the risk of delegated control becomes difficult to see. AI rarely takes control by issuing an order. It gains influence by becoming the easiest source of the next sensible step.

The machine does not need to force us.

Convenience can do the work.

A 2025 Microsoft Research and Carnegie Mellon study surveyed 319 knowledge workers and collected 936 real examples of generative AI use at work. The researchers found that higher confidence in AI was associated with less critical-thinking effort, while greater confidence in one’s own ability was associated with more.

But the finding is more nuanced than “AI makes us stop thinking.”

The researchers also found that critical thinking changes shape. It moves toward verifying information, integrating AI output and overseeing the task.

That is encouraging.

It suggests the answer is not to reject AI. It is to redesign our relationship with it.

• The goal should not be maximum human effort.

• The goal should be conscious human authorship.

Offloading is not the enemy

We should be careful not to romanticise mental effort.

Nobody becomes more human by refusing a calculator and doing long division for pleasure. I do not need to memorise every phone number I once knew. I am happy for GPS to get me through an unfamiliar city.

Cognitive offloading is normal.

A 2025 study in the journal Societies, involving 666 valid participants, found an association between heavier AI-tool use, greater cognitive offloading and lower critical-thinking performance. The authors also reported that younger participants showed higher AI reliance and lower critical-thinking scores than older groups.

That finding deserves attention, but not panic. The study is correlational. It does not prove that AI causes cognitive decline.

It does point to a question we should take seriously:

If a capability is repeatedly outsourced, what happens to our ability or willingness to exercise it ourselves?

That question matters most for capacities that are not merely mechanical as memory can be supported by a machine.

But what about:

• Judgment?

• Taste?

• Moral responsibility?

• Identity?

• Choosing a life direction?

• Knowing when a persuasive answer is wrong?

These are harder to delegate because the “correct” answer cannot be separated from the person who has to live with it.

The thing to protect is authorship

I think we often frame the AI debate incorrectly. We ask whether humans will remain “in the loop.”

But being in the loop can mean little more than clicking “Approve”.

That is not agency.

The better question is: Did the human meaningfully author the decision?

Authorship does not mean doing everything yourself. An author uses editors, researchers, tools, conversations and evidence. The work can be collaborative.

But the author remains responsible for what finally appears on the page.

That is the relationship I want with AI.

• Not servant and master.

• Not oracle and follower.

More like a powerful intellectual collaborator whose work I remain responsible for accepting, rejecting or changing.

That requires a boundary.

The machine can widen the field of possibility. But it should not quietly become the owner of my preference.

Five rules for retaining agency

Human agency is where we feel like we have control. And in a world of AI it can be slowly taken from you without noticing. But you can keep control by following some simple rules

1. Choose before you ask. When the decision matters, form an initial view before prompting AI. It can be incomplete and wrong. The point is to preserve your starting signal before the machine supplies a polished alternative.

2. Ask for opposition, not agreement. Do not only ask, “What should I do?” Ask, “What assumption am I making? What is the strongest argument against this? What evidence would change my mind?” AI becomes much more valuable when it makes your thinking harder to fool.

3. Separate recommendation from decision. Make the boundary explicit. The machine can research, rank and recommend. Name the moment where responsibility returns to you. For trivial decisions, this may not matter. For consequential ones, it does.

4. Keep some useful friction. Friction is not always inefficiency. Sometimes the pause is where judgment appears. If the machine can draft the message instantly, spend another minute asking whether the message should be sent. If it can generate ten strategies, resist asking for twenty. Choose.

5. Decide what you will not delegate. This may be the most important rule. Make a short list of decisions that require your conscious authorship: who you become, what you value, which relationships matter, what work deserves your time, what risks are worth taking, and what promises you will make. AI can help you think about these questions. It should not answer them on your behalf.

The next frontier is not smarter machines

The easy response to all this is to say that humans have always adapted to new technology.

And we have and we will again. But adaptation is not automatically flourishing. Technology always changes what we can do.

The deeper question is what repeated use changes in us as AI will become:

• More convenient.

• More personalised.

• More persuasive.

• More capable of anticipating what we might want before we ask.

That is precisely why the boundary needs to become clearer, not weaker.

The future I want is not one where humans stubbornly refuse help from machines. It is one where we use extraordinary machine intelligence to become more capable of choosing, creating, connecting and acting in the real world.

• The machine can help us see.

• It can help us think.

• It can help us move.

But at some point it has to hand the decision back.

“Because perhaps the great challenge of the AI age is not keeping humans in control of machines. It is keeping humans consciously in control of themselves.”

If you want to keep exploring how we use AI without surrendering what makes us distinctly human, subscribe to Human Signal Lab.

From thinking about agency to practising it

One reason I am building Zyrro is precisely this problem. AI can give us more possibilities than we know what to do with. Zyrro is being designed to start somewhere different: with who you are, what matters to you, and what you consciously choose to act on.

The aim is not to let AI decide your life.

It is to use AI to help you understand yourself, make better choices and act on them.

Join the Zyrro waitlist →

Sources and further reading

PNAS Nexus (2025): Engagement, user satisfaction, and the amplification of divisive content on social media — Preregistered audit comparing Twitter’s engagement-ranked and reverse-chronological feeds.

Microsoft Research / CHI 2025: The Impact of Generative AI on Critical Thinking — Survey of 319 knowledge workers across 936 real-world uses of generative AI.

Societies (2025): AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking — Mixed-method study with 666 valid participants; interpreted cautiously because its key relationships are correlational.

Zyrro — Waitlist and overview.

Publishing note: The two grey subscribe-box markers are intentional place

There is a road you have driven a hundred times.

Try to draw it from memory. The turns. The order of the streets. The way the light falls at the third intersection.

Most of us can’t.

Researchers at McGill tested 50 regular drivers and found something worth sitting with. People with more lifetime GPS use had worse spatial memory when they had to find their own way without it. Three years later, the heavier users had declined faster. And they had not started out with a poor sense of direction — the GPS came first, and the decline came after (Dahmani & Bohbot, Scientific Reports).

Nobody handed over their sense of direction.

They just took the fastest route. Every time. For ten years.

One of the researchers, Véronique Bohbot, has a blunt prescription. Getting lost, she says, pulls you out of habit mode and switches the brain back on (McGill).

Now…let’s change the subject from directions to decisions.

Subscribe now

The debt you don’t feel

Last year MIT’s Media Lab put EEG caps on 54 people and had them write essays over four months. One group used ChatGPT. One used a search engine. One used nothing but their own head (Your Brain on ChatGPT, MIT Media Lab).

The brain-only group showed the strongest, most widely distributed neural networks. The search group sat in the middle. The AI group showed the weakest connectivity of all.

A day later, they were asked to recall the text they had just written.

Chart 1 — Recall of your own words, 24 hours after writing. Source: Kosmyna et al., MIT Media Lab (2025)

The researchers gave it a name that should make every one of us pause: cognitive debt.

It is a small study. It has not been through peer review. Hold it lightly.

But hold it.

Because a second piece of research points the same way from a different angle. Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real tasks. The more confidence people had in the AI, the less critical thinking they did. The more confidence they had in themselves, the more they did (Lee et al., CHI 2025).

The tool is not the variable. The ratio is.

Two kinds of confidence sit on a seesaw, and every time you trust the machine a little more, you trust yourself a little less.

That is not a technology problem. That is an identity problem wearing a technology costume.

Nobody owns the decision

Here is where it gets interesting.

The danger is not that AI makes bad decisions for you.

The danger is that it makes good ones.

Bad advice teaches you to argue. Bad advice keeps your hand on the wheel. You read it, something in you objects, and the objection is the muscle working.

Good advice teaches you to stop objecting.

And that is the whole story. Dependence does not arrive as a decision. It arrives as relief.

It moves in four steps, and none of them feel like surrender.

Chart 2 — The adoption ladder. Convenience to infrastructure, one comfortable step at a time. Model: The Human Signal Lab

The villain in this story is not a company or a model or a founder in a black t-shirt. The villain is the default. A setting somebody else chose, that you never noticed, that now runs the part of your life you assumed you were running.

Nobody signs anything. That is the trick.

Subscribe now

The cockpit

We have run this experiment before. At 35,000 feet.

Airline pilots hand-fly on takeoff and landing. The machine has the aircraft roughly 90% of the time (US DOT Inspector General). An expert panel commissioned by the FAA warned that pilots were losing basic flying skills and named the condition “automation addiction” (NBC News). In one FAA research review, eroded skills were a factor in around 60% of the accidents examined (AVweb).

Chart 3 — The machine flies. The pilot watches. Then the machine disconnects. Sources: US DOT OIG; FAA flight deck automation panel

Now the part nobody wants to say out loud.

Automation made flying the safest way to travel in human history. The same systems that thinned the pilots’ skills also saved thousands of lives. Both things are true in the same sentence.

That is the tension, and I am not going to resolve it for you, because it doesn’t resolve.

I will own my side of it.

I use AI every day. It carries my research. It builds my charts. It stress-tests my structure and finds the hole in my argument at 5am when there is nobody else awake to tell me I’m wrong.

The headline is mine. The story is mine. The opinion is mine.

But I feel the pull. There are mornings when the machine offers me a finished thought, and the honest thing to admit is that taking it would be easier than having one.

The decision no machine would have approved

In 2009 my business was gone and I was 52 years old.

I started getting up at 4:30 in the morning to write a blog. The first post was about Jennifer Aniston, John Mayer and a website called Twitter that almost nobody I knew was using.

Feed that situation to any optimizer, then or now. A man in his fifties. No audience. No product. A collapsed business. A medium with no proven path to money.

The answer comes back in half a second. Get a job. Update the résumé. Reduce risk.

Seventeen years and 33 million readers later, I am 69 and I have started an AI company.

I posted that fact on LinkedIn once and asked if I was mad. The post went nowhere. Which is its own kind of answer.

Here is what I have worked out, and it took me most of my life.

Every decision that changed me would have failed an expected value test.

Leaving the church at 27. Leaving teaching. Writing into an empty room at dawn. Building a product at an age when most people are being told to wind down.

An optimizer runs on the sentence and the point that cracks things open!

“It cannot know what would work if you became a different person, because the version of you it is modelling has never met the version of you that decides”.

That is what is at stake.

Not productivity.

Not output.

Not the hours you save.

The capacity to make a choice your own history does not support.

Subscribe now

Tie yourself to the mast

Odysseus wanted to hear the sirens.

He did not plug his ears and sail past with the crew. He wanted the song — all of it — so he had himself lashed to the mast and gave one order in advance: whatever I scream at you out there, ignore me.

He made one decision on dry land so he would not have to make a hundred decisions while under the spell.

That is “agency” in 2026.

Not refusing the machine. Not romantic abstinence, not writing your emails on a typewriter to prove a point.

Deciding, before you are in the water, which choices are yours.

Because the water is rising. 62% of Australian consumers say they are open to AI agents making purchases for them — household items, subscriptions, gifts, travel. Morgan Stanley expects close to half of online shoppers to be using AI shopping agents by 2030, covering around a quarter of their spending (commercetools).

Chart 4 — Discovery, choice and purchase are already moving to the agent. Sources: commercetools (2026); Morgan Stanley forecast

It starts with toothpaste.

It does not stop at toothpaste. It moves to who you hire, where you live, what you charge, what you write, who you become.

The line

So where is the line between help and handover?

I have three tests. They are not a framework. They are what I use at my desk.

Does it hand the decision back, or keep it? A good tool returns the choice to you with better light on it. A bad one returns a conclusion and hopes you don’t look up. Watch which one you reach for when you are tired.

Afterwards, do you know more about your own reasons, or less? Cognitive debt is invisible in the moment, because the work is finished and the output is fine. The tell is not the quality of the answer. The tell is whether you could defend it tomorrow to someone who disagreed.

Delegate the how. Keep the why. Research, formatting, first drafts, the eighteen tabs, the spreadsheet — take all the help you can get. Purpose, taste, the thing you are willing to be wrong about in public — those are not tasks. They are the parts of you that make the work worth doing at all.

And build in some friction on purpose. Bohbot’s advice for GPS was to use it on the way somewhere new, and turn it off on the way home.

Same rule. Use the machine to reach the unfamiliar. Find your own way back.

The verdict

I am not worried that AI will outthink us.

I am worried that we will stop noticing when it thinks for us — that judgment will thin the way a pilot’s hands go soft after a decade of watching the autopilot do the landing, and that we will only find out in the one moment that mattered, when the system disconnects and the aircraft is ours again.

Choosing is not a skill you keep by having it. It is a skill you keep by using it.

So use it on something small this week. Something the machine could have settled for you in four seconds. Sit in the discomfort of not knowing, long enough to find out what you actually think.

There is a road you have driven a hundred times.

Drive it without the voice.

Get lost on purpose. It is the cheapest way to find out you are still the one at the wheel.

BUILT TO MAKE YOU NEED IT LESS

I am building Zyrro because of everything above.

Most AI is designed to answer for you. Zyrro is designed to hand the decision back, a reflective intelligence platform that helps you see your own patterns, name your direction and act on it. It starts with a free ten-question Signature Read.

The measure of success is not how much you use it. It is how much of your own judgment you get back. Join the Zyrro waitlist at zyrro.ai

Sources

• Dahmani & Bohbot, Habitual use of GPS negatively impacts spatial memory, Scientific Reports (2020) — read

• Kosmyna et al., Your Brain on ChatGPT: Accumulation of Cognitive Debt, MIT Media Lab (2025 preprint) — read

• Lee et al., The Impact of Generative AI on Critical Thinking, Microsoft Research & Carnegie Mellon, CHI 2025 — read

• US DOT Office of Inspector General, on pilot manual flying skills and automation — read

• FAA expert panel, flight deck automation and “automation addiction” — read

• FAA research review, skill erosion as a factor in reviewed accidents — read

• commercetools, Agentic commerce statistics and Morgan Stanley forecast (2026) — read

Computer algorithms that decide who is seen and what content is shown to everyone online are built into every platform. They are the “black boxes” of the Web. No one quite knows what they do and how they do it.

But they decide who is important and what is important.

It was simple once as most people knew what the rules of the game were. But what’s happened since is a dilemma that dogs, fogs and confounds me.

I started on Twitter, Facebook and search engines when the algorithms were simple.

Everyone who played was more noticeable because they knew the rules of the game.

Able to be watched. And I could dance with the machines. It felt like I had a chance to win.

In 2026, the power of AI is built into everything. The machine has the rules built in. And I am starting to feel almost powerless. As the rules are invisible.

The AI algorithms determine what’s visible.

• To be shown in the feed of the social media machine

• To be found on a search engine.

• To be visible as a personal brand.

• To build a business that works online.

Unless you pay.

So my thought is this and questions that rise:

• Is this just a frustration I sense?

• Does more effort no longer work?

• Is persistence futile?

• Does the research back it up?

• Should I pack up and give up?

So I stopped guessing. And I went looking for the evidence. Here is what I found.

A box that records everything and shows you nothing

There is a black box on every plane. It records the whole flight. Every sound. Every choice. And you never get to open it. Unless you crash.

The web now runs on a box like that. It watches what you make. It reads every word you write. Then it decides who gets seen. And it never tells you how.

You used to be able to study the machine and learn its moves. Not anymore. The box got smarter. And it went quiet. You didn’t crash but you got buried.

I remember the old dance. You watched what spread. You spotted a pattern. You leaned in, and the machine rewarded you. It was a game two could play.

Now only one side can see the board. The platforms. Nobody voted for this. It just became normal, one small default at a time.

And here is the part that should sting. You helped build it. Every post fed it. Every article trained it. We handed the box our life’s work, one upload at a time, and it never signed a thing.

They kept the click

For years the deal was simple. You write something good. You show up in search. The reader clicks. The reader arrives.

That deal is breaking.

Across 68,879 real Google searches, when an AI answer sat on the page, clicks to a website fell by nearly half.

Source: Pew Research Center, 2025

The wider count is worse. Out of every 1,000 Google searches, only about 360 now send a person to the open web.

Six hundred and forty people ask a question. The machine answers. The door to your work stays shut.

You still win the argument.

The reader still never arrives.

The reach that built us is almost gone

Once, a page you built could reach the people who followed it. That was the whole promise of social media. Post, and be seen.

In 2012 a Facebook page reached about one in six followers. Today it is closer to one in twenty-five. LinkedIn reach has dropped about a third.

Source: Social Media Examiner / industry data, 2026

You did not get worse.

The house changed the odds.

It is like renovating a rented flat. You make it beautiful. Then the landlord changes the locks.

The real cost of the box

Here is the number almost nobody shows you.

Cloudflare measured how many pages each company takes from the web for every single visitor it sends back.

The old search bargain: five pages taken for one visitor returned. The new one runs to hundreds, then thousands, for one.

Source: Cloudflare Radar, May 2026

The old search deal was five pages taken for one visitor returned. A fair trade.

The new deal is different.

For OpenAI, it is more than eight hundred pages for one visitor.

For Anthropic — the maker of the very assistant helping me research this — it is over eleven thousand.

Picture a shop.

A machine walks in. It photographs every item on every shelf, eleven thousand times over. Then it buys one thing and leaves.

That is the box.

It takes almost everything.

It gives back almost nothing.

And what does it do with what it takes? It learns from it. You spend ten years finding your own way of saying things. The box copies that style in an afternoon, then hands it to a stranger for free, with your name stripped off.

The money is climbing toward the top

And building a business on top of all this? The ground is tilting.

The top 1% of creators now take 21% of the money, up from 15% two years ago. The middle is being squeezed out.

Source: CreatorIQ State of Creator Compensation, 2026

The top 1% of creators now take 21% of the money. Two years ago it was 15%. The top 10% take 62%.

The middle is vanishing. The writer with 20,000 real readers who used to pay the rent from her work. The steady voice with a small, loyal following. They are the ones losing ground.

This is where “unless you pay” comes from. It feels like a toll booth. Pay, or stay invisible.

Hold that thought. Because here the story turns.

The surprise

I blamed the slop.

The flood of cheap, machine-made words. I said it was burying us. Stealing our light.

Then I read the data.

And the data says I was wrong.

Half of all new web articles are now made mostly by AI. But only 14% of the pages that win the top spots are.

Source: Graphite / Common Crawl, 2026

Read that again.

The flood fills the web. It does not fill the winners’ circle.

The slop is drowning in itself.

And you cannot pay your way past a power law.

The top 1% did not buy their seats. No cheque moves you there.

So the villain is not the toll booth.

The villain is simpler, and worse. The kind of writing most people sell has just become free.

Think of the plain how-to post. “Seven ways to be more productive.” “A beginner’s guide to email marketing.” The box writes those in seconds, at no cost. So that kind of writing is now worth close to nothing. Anyone can make it. Everyone does.

One machine can make the average thing for free. So the average thing is worth nothing. That is the whole story in one line.

What the box cannot copy

I have had the same cleaner for fifteen years.

She knows my history. I know hers. She has watched my family grow.

She is part of the family. I could never swap her for a robot.

Why not? A robot can do the task.

Because it was never only a task. It was a relationship.

The box can handle a task.

It cannot hold a relationship.

The box can copy your words. It cannot feel them.

A robot can do a job. It cannot be part of a family.

That same line runs straight through your writing. The box can copy what you know. It can never copy who you are.

The reader is starving

Now turn it around. Look at the person on the other side of the screen.

They are drowning.

A feed of sameness. A search page of summaries. A thousand posts that sound like one machine wrote them — because it did.

Somewhere in that flood sits one person in a small office. Scrolling. Thinking the same three words you once thought.

There has to be more.

They are not starving for more content. They are choking on it.

They are starving for a human. A real one. A voice with a pulse.

Not a brand. Not a logo. Not a feed. A person who has lived something and is willing to say it out loud.

That is your opening. That is your whole business, right there.

The web is full of copies now. One real person, saying one true thing, is the rare one in the room. That person can be you.

How you win anyway

Five hundred years ago the printing press made words cheap. The scribes panicked. Their craft was copying, and copying was now free.

But something strange happened.

When words became cheap, the writer became precious. The named human. The voice. The one you trusted.

The machine has done the same thing again. Words are cheap now. So the person behind them is the rare and valuable part.

So here is how you win against a machine web.

Not by feeding it more.

For years the answer to every drop in reach was the same. Work harder. Post more. Publish faster. Feed the machine what it wants.

That answer is dead. You cannot out-work a machine that never sleeps and never asks to be paid. Every hour you spend making another me-too post — another “ultimate guide” the box could write for free — is an hour you set on fire.

Effort pays now in one place only. On the things the machine cannot make. Your own story. The client you actually helped, and how. The mistake you actually made, and what it cost you. The opinion you are a little scared to publish.

You win by owning the ground under your feet. Look at Substack. More than eight million people now pay for writing there, and writers earned over $450 million last year. Not one of them bought their way in. They own the door. No landlord. No locks.

An email list is the one asset on this whole web that no algorithm can throttle and no landlord can seize. You built it. You keep it. It is yours.

You win by being the voice people trust, not the pile of words a machine skims. When anything can be faked, people stop trusting polish. They start trusting proof that a real person was there. So show your working. Name the real client. Post the ugly first draft next to the finished one. Say “I got this wrong last year, and here is what it taught me.” A machine cannot do that. It was not there. You were.

You win by being recommended, not ranked. Ranked is being result number four on a page nobody scrolls. Recommended is being the person a friend texts about: “you have to read this.” One of those you fight the box for. The other, the box cannot touch.

You win by trading reach for depth. One reader who opens your email every week is worth more than a hundred strangers who scroll past a post and forget it by lunch.

You win by counting the right things. Not clicks. Not follower counts. Not the numbers the box wants you chasing. Count the people who would notice, and care, if you went quiet for a month.

And you win by being no one but yourself. Do not write “five tips for better emails.” Write “the email that got me fired, and the three things it taught me.” The first, a machine can write in a blink. The second, only you can, because only you lived it.

Here is the catch. Knowing which story is yours to tell — the point of view only you carry — is the hard part. That is the problem I am building Zyrro to solve. It reads your patterns back to you and shows you the signal underneath your work.

Step onto the Zyrro waitlist →

See what runs through all of it? Every move points the same way. Away from what the machine does well — cheap, fast, average. Toward what it cannot do at all. Be a person.

The machine scales your output. It cannot scale your soul.

You were never the customer

Let me stop and tell you what really happened. Plainly. Because once you see it, you cannot unsee it.

A platform needs two things to get rich. A huge crowd of people. And a steady flood of content to keep that crowd staring at the screen.

Making that content costs a fortune. So the platforms found a cheaper way. They got you to make it. For free.

Here is how the trick worked.

In the early days, they gave reach away. You posted, and thousands saw it. No charge. It felt like a gift.

So you kept posting. More articles. More videos. Your best thinking, poured onto their platform, for nothing. Millions of people did the same.

All that free work is what made the platform worth billions. Your content was the bait that pulled the crowd in. And the crowd was the thing they sold to advertisers.

You thought you were the customer. You were not. You were the free labour that built the whole thing.

Then the building was finished.

Once everyone had moved in, and the platform had all the content and all the attention it would ever need, it stopped needing you. So it turned the free reach off. Quietly. No warning. No email.

The reach that was free now cost money. Pay to be seen, or disappear.

You were not the partner. You were not even the product. You were the scaffolding. And once the building stood on its own, they took the scaffolding down.

That is the whole con, in one line. They used your work to build their business. Then, once it was built, you became collateral damage.

The same trick, three times

And here is the part that should stop you cold. This is not the first time. It is the third.

The first time was Facebook. It handed out free reach until hundreds of thousands of writers and businesses had made it their home. Then, in 2012, when it went public and had to make money for its new owners, it choked that reach off. A post that once reached a hundred thousand people now reached a few hundred. Overnight, and there was nothing you could do.

The second time was Google. For twenty years it sent you free visitors, until the whole world wrote its websites to please it. Then Google began answering questions itself, right there on the results page, and kept the visitor for itself. Your work fills the answer. Nobody arrives at your door.

The third time is now. It is AI, and it is the largest grab of all. The machines read almost everything you ever wrote — for free, without asking — and used it to build tools that now do your job. They took a lifetime of your work as raw material, and hand you back almost nothing.

Same trick, three times. Dangle something free. Feed on the work you give them. Then, once they have won, cut you off, or replace you.

You survived the first two. You will survive this one.

But not by playing their game. You cannot win a game where they own the board and change the rules whenever it suits them.

The box is real. The battle is real. The research says so.

But the box has one blind spot, and it is the size of a human being.

It can take your words.

It can rank your page.

It can copy your style.

It can never be you.

So stop trying to beat the box at its own game. You cannot. It never sleeps, and it works for free.

Play the game it cannot play at all.

Tell the truth. Sign your name. Be the person it can copy but never become.

Be the signal.

Finding your own signal is the work of a lifetime. It is also the one thing I am building Zyrro to help you do. If that speaks to you, the early list is open.

Join the Zyrro waitlist →

I had to Google the word “SEO” ten times before I understood it.

This was 2008. I was fifty-one. Something was forming on the internet and it had grown its own language before most people noticed it existed.

Search engine optimisation. Content marketing. Personal branding. Social media marketing. Foreign words. New words. Nobody was fluent, including the people using them.

So I did the only thing I knew how to do. I took notes in public.

Short posts. Tweets, Observations.

Whatever caught my eye. I was not teaching anyone. I was trying to work out what was happening, and writing was how I thought.

Fourteen years and thirty-three million readers later, I had a media company built on top of that language.

Every word of it is now obsolete.

Not wrong. Obsolete. The habits still fire, the keyword in the headline, the publishing schedule, the impressions checked before breakfast — but the machine they were feeding has been switched off underneath them.

And here is the part that keeps me awake.

I did not just learn that language. I taught it.

The villain is not a technology. It is a deal.

Everyone wants to point at AI. That is the easy answer and it is the wrong one.

AI is the third repossession.

1. In 2012 I had a Facebook page that reached the people who had chosen to follow it. Then organic reach was throttled, and reaching my own audience started costing money. That was the first.

2. Then Google began answering the question on the results page. The snippet took the answer. The click stayed home. That was the second.

3. Now the answer engine takes it entirely. That is the third.

Three separate events. Fifteen years. One identical arrangement underneath all of them.

You build the audience. Someone else owns the road.

That is the villain. Not a technology but a deal. And we have signed it three times, each time convinced the new road was different from the last one.

“AI broke my business” is a complaint.

Subscribe now

“I rented my audience three times and was evicted three times” is a confession. And it is the true one.

What the third eviction looks like

The Pew Research Center watched 900 people run 68,879 Google searches. When Google placed an AI summary at the top of the page, 8% clicked through to a website. Without a summary, 15% did. Links inside the summary itself were clicked one time in a hundred.

SparkToro then measured the whole market with Similarweb’s panel. In the first four months of 2026, 68% of US Google searches ended without a click to anywhere. Not to a website. Not to an ad. Not even to one of Google’s own pages.

Chart: The Vanishing Click. Source: SparkToro / Similarweb, June 2026

Ten years ago that figure was around 45%.

Here is the version that hurts. For every 1,000 searches, 276 clicks now reach the open web. Two years ago it was 374.

And it is not evenly spread. Ten major tech and media sites saw their combined US Google traffic fall from 112 million monthly visits to under 50 million. Digital Trends went from 8.5 million clicks a month to 264,861. A fall of 97%.

The first thing that broke was the answer

I have written hundreds of posts called “The Ultimate Guide To Something.” For years they worked. They paid the mortgage.

Now look at the largest answer library human beings ever built.

Chart: The Answer Business. Source: Stack Exchange data, via PPC Land

Chegg sold homework answers by subscription. Its share price fell 99% over five years.

Same disease. Both were in the business of storing answers somewhere they could be found.

If the value was the answer, it is gone.

If the value was the person who had to live through something to know it, it is not.

Machines hand out answers. They cannot hand out scars.

The second thing was volume

Publish more, get found more. It was the only strategy anyone offered for fifteen years. I taught that too.

Graphite sampled tens of thousands of pages from the open web. Machine-written articles now make up more than half of everything newly published.

Now the part almost nobody quotes, from the same researchers.

Those machine-written articles largely do not appear in Google results or ChatGPT answers at all.

Infinite supply. No visibility.

We flooded the world with words that nothing will ever read.

The third thing was proof

You sent forty applications and heard nothing back. You assume something is wrong with you.

Subscribe now

LinkedIn now takes around 11,000 job applications every minute, 45% more than a year earlier. That is not more people wanting work. It is the same people, with AI firing off hundreds while they sleep. The average opening now draws 242 applications.

And 19% of organisations using AI to screen candidates have confirmed the software rejected people who were qualified. Not suspected. Confirmed.

Two machines are holding a conversation about your life and neither of them has met you.

Google and Cisco have started interviewing in person again. In 2026, the most reliable way to prove you are a human being is to stand in a room beside another one.

The sentence I did not want to write

Harvard, Imperial College London and the German Institute for Economic Research went through two million freelance job posts across 61 countries. Within eight months of ChatGPT arriving, demand for freelance writing fell 30%. The steepest fall of any category they measured.

Then Washington University tracked 92,547 writers and found the pattern that should stop all of us. The best paid and highest rated were hit hardest.

Quality was never the moat. It was just expensive to copy.

Now let me argue against myself

I have staked a lot on the idea that the human signal is what survives. That makes me exactly the man most likely to see collapse everywhere and call it proof. So here is the case against my own article.

• The flood had a ceiling. In 2022, Europol predicted 90% of online content would be machine-made by 2026. It reached 50.9%, stopped, and drifted back to 49.9%. People kept writing.

• Traffic moved more than it vanished. Chartbeat’s network of more than 2,500 publishers lost 34% of its Google search traffic in a year, but total pageviews fell only about 6%.

• Some names went up. In the same period, the Guardian and the BBC gained. They did not out-optimise anyone. People typed their name.

• The money did not leave. Fiverr lost buyers last year while spend per buyer rose 8.3%. Fewer clients. Bigger cheques.

None of that says nothing broke. It says something more precise, and far more useful.

It is not a collapse. It is a barbell.

At one end, anything cheap and generic went free.

At the other end, anything rare and named got more expensive.

The middle fell out.

And the middle is where almost all of us built our careers. Good enough. Reliable. Competitive on price. Findable.

Upwork writing projects dropped 32% in a year while specialists went the other way. Finance writers averaging around $73,000. Some fintech work paying 95 cents a word.

Same market. Two directions. Look down at your feet and work out which end you are standing on.

Everyone is making the 2009 bet again

Here is what worries me most, and almost nobody is saying it.

Watch what people are doing right now. Learning GEO. AEO. Answer engine optimisation. Rushing to master a fresh set of acronyms in exactly the way I mastered the old ones.

I got fifteen years of return on learning SEO. The vocabulary and its tactics that worked held long enough to build a company on.

Nobody gets fifteen years now.

It is more like two years.

The half-life of a platform tactic is under two years and shortening. Anyone going deep on the new language is buying a skill that expires before it compounds.

So the adaptation is not learning the new vocabulary.

It is learning it shallowly, on purpose.

Rent the tactics. Own the thing underneath.

Then….

Test and experiment ad iterate at high velocity

What survived was the method, not the output

Look at what I actually did in 2008. I did not learn SEO. I noticed a change before it had a name, wrote my way into understanding it in public, and became the person who could explain it.

The output died. The method did not.

It is the only thing that has survived all three evictions, and the only thing that was ever mine.

Four moves follow from that. All testable this month.

1. Publish only what cannot be regenerated - Primary observation. First-party data. What happened to you last Tuesday. If a machine could produce it from the public record, you have not written anything. You have filled a slot in a system that no longer pays.

2. Replace impressions with a name count - Branded search. Direct traffic. People who arrived because they were looking for you specifically. It is a far smaller number and it is the only honest one you have.

3. Sell judgment, not output - Output is now priced at machine cost, which is zero. Judgment is what you do with output, and it stays scarce because it takes a life to build.

4. Shorten the road to zero intermediaries - One list. One room. One group of people who would notice if you stopped. Not as insurance. As the primary asset. Everything else is rented, and you already know how that ends.

This is the work I am doing now. Zyrro starts with who you are rather than what you produce, because identity is the one thing on this list that did not get cheaper. If the four moves above are the practice, that is the place I am building to do them in. zyrro.ai

The verdict

Nothing on the broken list was ever the real asset. Rankings, volume, credentials, follower counts. Every one of them was a machine for proving something about a person, built in an age when proof was expensive.

Proof is free now. It is worth what free things are worth. What is left is the person. Not the summary of the person. Not the profile. Not the searchable record of what they once did.

The one who had to live something in order to know it.

The curse is that we will probably do it again. There is a new road being built and it looks nothing like the last three, which is precisely what we said the last three times.

The opportunity is that for the first time in twenty years, the cost of not renting has fallen to almost nothing.

I had to Google “SEO” ten times before I understood it.

I am not going to spend the next fifteen years learning a language somebody else owns.

Two things, if this landed.

Reply and tell me which of the four moves you are testing this month. I read every one of them.

And if you would rather work on the thing underneath than the tactics on top, that is what I am building at zyrro.ai.

In the last two years, the web changed the rules on who gets seen. Here is what happened and the part of you it’s been hiding.

You made something.

You put it into the world. And the world did not look up.

Not because it was bad. You know it wasn’t bad. You sat with it. You cut the weak lines. You pressed publish with that small, private hope.

And it landed in front of almost no one.

You blamed the topic. Then the timing. Then the headline.

It was none of those.

Somewhere in the last two years, the whole web changed the rules on who gets seen. And it forgot to send any of us the memo.

There is one word that keeps coming back for this. Not unpopular. Not rejected.

Invisible.

Seen by no one, for reasons no one explained.

This is the story of why. And what it is quietly asking you to become.

It is happening everywhere at once

This is not one app’s problem. It is the shape of the whole web now.

On X. On Instagram. On YouTube. On LinkedIn. On TikTok, where it began. It is creeping into search, where an AI answers the question before anyone clicks your link.

The whole web is being re-sorted.

Not by who you know. By what a machine decides you are.

The day the deal changed

For years the web ran on the social graph.

The social graph asked one question. Who do you know?

You followed people. People followed you. You built trust over years, one relationship at a time. When you posted, your followers saw it. That was the deal.

Then TikTok tore up the deal.

TikTok did not care who you knew. A stranger with no followers could reach a million people by lunch. The machine watched what you watched, decided what it was about, and served it to people who liked that sort of thing.

They call it the interest graph.

The interest graph asks a different question. Not who do you know. What are you about?

And it worked so well that everyone copied it.

The machine took the door

Facebook went first. Back in 2014, a study of more than 100 pages found reach falling from about 16% of followers to around 6% in two years. For the biggest pages it hit 2%.

They called the report Facebook Zero. We did not want to believe it.

Then it climbed inside the feed itself. On earnings calls, the man who runs Facebook and Instagram has tracked how much of what you see is now chosen by AI from accounts you do not follow. In 2022 it was about 15%. By 2024, more than half of everything on Instagram.

X is the same. Its own engineers say the For You feed is about half from people you follow and half from a map of roughly 145,000 interest groups the machine drew on its own, redrawn every three weeks.

The professional networks work the same way, in a suit.

Their engineers have published how the feed decides who sees your post. It first turns you into words a machine can read. Your behaviour. Your history. Your connections. All of it rewritten as a description.

Read that again.

To decide if you are worth showing, the machine first writes a description of you.

Now the part the critics leave out

This is the best thing that ever happened to the unknown.

A person with no audience, no network, no ten years of showing up can be heard tomorrow. A good idea from a nobody now stands next to a somebody. That is new. That is fairer than the old web ever was.

Because the follower era was never the clean race we remember. It rewarded arriving early and posting often. I know. I started a blog in 2009 with ten dollars and no audience, at 52, with a failed business behind me. Some of my reach was a prize for turning up first, not for being right.

Your followers were never yours.

They were rented. You just did not have the word for it.

So there is no villain here with a face. The villain is a mechanism. And it means well.

What the machine asks of you

To recommend you, the machine needs a clean answer to one question.

What is this, and who is it for?

That is fair. It is also a demand.

Subscribe now

Be easy to file.

People who study this have watched what happens to those who are not. A firm that does not sit inside one clear category gets covered by fewer experts and priced at a discount. Films that mix too many genres score lower with audiences. Freelancers who leap between unrelated jobs get read as a dabbler, not a Renaissance mind.

One honest note. When someone tried to repeat the first of those studies years later, two of its three results did not hold up. The penalty for being hard to file is real in some places and shaky in others. That is what a fair reading looks like. And it matters more than a clean story.

What a music service found out about its own listeners

The strongest proof I found came from inside a company.

In 2020, researchers at a music service studied more than 100 million people and around 70 billion plays. They scored every listener from narrow to broad.

Three things fell out.

Listening picked by the algorithm was nearly always narrower than listening people chose for themselves.

When people grew broader over time, they did it by walking away from the algorithm and back to their own choices.

And broad listeners were far more likely to stay, and far more likely to pay.

The company’s own scientists found that the listeners the machine serves best are the ones most likely to walk out the door.

The machine is not broken. It is short-sighted. It is brilliant at the next three minutes and weak at the next three years.

The quietest kind of censorship

There is a second force at work, and this one lives inside you.

A researcher at Duke ran six experiments on what happens when you measure something you enjoy. Measuring made people do more of it — and enjoy it less. Watching the number turned play into work.

Nobody told them to stop. A number did.

Now think about you. Every post hands back a score within the hour. Views. Saves. New followers.

Nobody censors you. You just look at the shape of what worked, and you lean toward it. Then you lean again. After two years of leaning you are a category, and you cannot remember choosing it.

That is the villain in this story. Not a company. A feedback loop with good intentions.

You now need three identities, not two

So where does this leave you?

It leaves you needing three identities.

The personal identity. Who you are when no one is watching. Your fears. Your history. The thing that gets you out of bed.

The public identity. What you choose to show the world. Your work, your voice, your face in the feed.

We have always had those two.

Now there is a third.

The AI identity. The version of you the machine builds to decide if you are worth showing at all.

You did not write it. You cannot see it. But it goes first.

Before a single human reads your words, a machine reads its description of you.

Miss the third, and the first two never get seen.

You become invisible.

Here’s the thing the machine will never tell you.

Under the category it filed you in, there is a fuller version of you. In your patterns. Your energy. The work that lights you up.

It’s hard to see on your own. That’s what I’m building with Zyrro — it reveals that fuller version, then helps you take the first real step toward it.

It’s live now, and the waiting list is open.

→ Become the person you’ve always sensed you could be. Get on the waiting list.

The trap inside the third identity

But here is where it turns.

If you flatten yourself into one neat category just to feed the machine, you win the wrong game.

Leonardo painted, cut open bodies, drew war machines, studied moving water, and spent years working out how birds stay up. Many subjects. One question. How does the world actually work?

Steve Jobs sat in on a class about beautiful lettering with no plan to use it. Ten years later it shaped the Mac. Zen, music, design, animation, shops. Many subjects. One question. How does technology become human?

A modern feed would have told Leonardo to stop posting anatomy, because people came for the paintings.

And the research sides with Leonardo, not the feed. A study of 17.9 million papers found the work with the most impact was mostly ordinary, with one strange spark thrown in. That odd mix was twice as likely to become a landmark.

Not narrow. Not random. A deep base with a strange spike.

What becomes scarce

Every wave of technology makes something cheap. And by doing so, it makes something else rare.

Machines made muscle cheap. The web made facts cheap. Feeds made reach cheap. AI has now made words and pictures cheap.

What is left is everything a category cannot hold. Your story. Your judgement. Your taste. What you have lived. Who trusts you. The gift of carrying an idea from one world into another and knowing why it belongs.

That is your human signal. And it is starting to look less like a style and more like a form of resistance.

Explore broadly. Publish coherently.

Not “be yourself.” That is a greeting card, not a plan.

Own the last step. An email list. A site. Somewhere they can reach you when the road closes. Roads are not land.

Choose a question, not a niche. Write it in one sentence. If it cannot hold five different subjects, it is too small.

Feed the third identity on purpose. Say plainly, again and again, what you are about — so the description the machine writes of you is one you would sign.

Explore broadly. Publish coherently. Wander in private. Land in public under one clear lens.

The file and the person

The machine will file you. That is its job. Most days it does it well.

But a file is a summary.

And no one in the history of the world was ever changed by a summary.

The machine already wrote its description of you. A summary. A category.

But there is a fuller version underneath it. The one you’ve always sensed you could be.

Zyrro helps you find it — and take the first real step toward it.

→ Find the fuller version of you. Get on the waiting list.

You weren’t born to be classified.

You were born to be found.

I have reinvented myself three times. I have never once taken a leap of faith.

I have observed what energises me and seen the patterns and chosen what my curiosity has demanded. It is an experiment of testing and watching to see what it reveals but also coupled with intuition to help me choose.

So…I’m building a company called Zyrro on exactly this pattern and more on that at the end. First, the three times it happened to me.

The story we are sold about changing your life has one shape. You are trapped. You wake up one morning, quit, burn the boats, and walk into the unknown with nothing but belief.

It makes a good film. It has ruined a lot of lives.

Here is what actually happened to me, three times, over forty years.

The staff room

I was 23 when I started teaching at a private high school in Melbourne. I was 27 when I left.

What ended it was not a bad day. It was a look around the staff room.

The teachers in their forties. Tired. Flat. Counting down to holidays and then to a pension. Good people who had stopped expecting anything to change.

I looked at them and understood I was looking at a photograph of myself in fifteen years.

Subscribe now

Nobody warns you about this. The danger in a career that does not fit is almost never failure. The danger is comfort setting around you like concrete while you are still young enough to move.

I also did not enjoy the work. My first career and six years of teaching people who had not asked to learn anything was a chore. They are called teenagers.

So I knew I had to go. I had no idea where.

That is the moment the leap story tells you to jump.

I did not jump. I already had evidence, and I had been collecting it for two years without realising what it was.

The tapes

The teaching pay was poor, so at night I sold subliminal suggestion tapes.

Stop smoking. Lose weight. Play the cassette while you sleep and the recording does the work.

Living room to living room. Kitchen table to kitchen table.

Some nights I earned more in three hours than I earned in a full week in the classroom.

The money was not the discovery.

The discovery was what those nights told me about myself. I was good at selling. I liked the pressure of it. I believed I was helping people change something they had failed to change alone.

Six years of teaching had told me none of that.

Here is the part I did not learn until decades later. In 1991, Anthony Greenwald and colleagues published double-blind tests of commercial subliminal self-help tapes in Psychological Science. They took 237 people who wanted the results. Some tapes were labelled memory, some self-esteem. For half the participants the researchers swapped the labels.

Neither tape produced the effect it claimed. But more than a third of people believed they had improved at whatever the label promised.

The product I sold was probably a placebo in a plastic case.

My customers still changed. Some of them stopped smoking. What changed them was deciding to, paying money for it, and telling another human being they had started.

I was not selling a recording. I was selling a decision with a ritual wrapped around it.

Hold that thought. We are about to be sold a great deal of AI the same way.

Three jobs, one summer

At the end of that year I had a long school holiday, so I used it as a laboratory.

I trialled three sales jobs. Real estate. Life insurance. Technology.

This was the early 1980s. The world was moving from mainframes and minicomputers to the personal computer. Jobs and Gates were about to rearrange everything, and almost nobody in Melbourne had noticed.

The money pointed at real estate. The security pointed at insurance. Something quieter pointed at technology.

I had a spreadsheet in my head and a whisper in my heart and they did not agree.

I followed the whisper. But I only trusted the whisper because I had spent two years and one summer gathering evidence that I could sell, that I liked selling, and that I could read a market before it arrived.

That is the whole method. The data earns you the right to trust the instinct.

I resigned in the new year. That choice put me inside the computer industry as it exploded, and everything since runs back through it.

It turns out this has a name

Thirty years later, two researchers at the University of Wisconsin gave my summer a label. Joseph Raffiee and Jie Feng studied more than five thousand people who started businesses, and published “Should I Quit My Day Job? A Hybrid Path to Entrepreneurship” in the Academy of Management Journal in 2014.

They compared two groups. People who quit and went straight into working for themselves. And hybrids, who built the thing at night while the job still paid the rent.

The hybrids who later went full-time survived at much higher rates. The figure most often quoted from the study is about a third more likely to still be trading.

The researchers pointed at some familiar names. Steve Wozniak stayed at Hewlett-Packard long after co-founding Apple. Pierre Omidyar launched eBay while working at a software company.

The reason is not complicated. Testing while employed lets you kill bad ideas cheaply. When the hybrid finally quits, they are quitting for something that already works.

Data: Raffiee and Feng, Academy of Management Journal, 2014

I did it again at 52

In my forties I opened a shop that sold beds. It went broke. And I was distraught. That experiment cost me over $250,000 and a marriage.

To pay the rent I took a job selling houses, which I was not good at. My marriage ended. I spent about a year at my brother’s place beside a lake, working out who I was going to be next.

Then in 2009, back in Sydney with a new partner and a spare room, I started a blog.

Not as a leap. As a test. The real estate job still paid for the groceries.

I got up at 4:30 in the morning and wrote, five days a week, for five years. The affirmation arrived in small pieces from strangers on other continents, and the motivation grew out of the action rather than coming before it.

It went on to reach 33 million readers.

Nobody would call that a leap of faith. It was a two-year experiment that quietly grew teeth.

And again at 69

This year I sold my house. I am debt free with what I call runway money.

I want to be honest about the part nobody posts. I was anxious making that decision. A house is not only an asset. It is a container you have built a life inside.

But it was the same move at a larger scale. I did not gamble. I had spent a year saving every book, article and paper that caught my attention, then asked an AI what I had been repeatedly trying to understand.

Five questions came back. Each one had been running in the background of my life for years.

One of them became Zyrro, the company I am now building.

If that sounds late, consider what Pierre Azoulay, Benjamin Jones, J. Daniel Kim and Javier Miranda found when they examined US Census records covering millions of company founders. In “Age and High-Growth Entrepreneurship” they report that the mean age of founders of the fastest-growing one in a thousand new ventures is 45.

Not 25. Not 30. Forty-five, and rising with industry experience.

The reason should be obvious by now. Older founders have more evidence about themselves. They have run more experiments, whether they called them that or not.

Why nobody tells you this

The leap makes a better story than the test.

Nobody films eighteen months of Tuesday nights. There is no dramatic scene in which a man carefully compares three sales jobs over a school holiday. The algorithm does not reward a decision that took two years to become obvious.

So the version that travels is the version with the cliff in it.

And a generation of capable people sit still, waiting to feel brave enough for a jump that was never required, in a career they already know is wrong.

The bravery was never the point. The evidence was.

How to run the test

This is the method I have used at 27, at 52 and at 69. It has not changed.

1. Keep the job. It funds the experiment and removes the panic that makes people take bad offers.

2. Run the real thing at night. Small, real, with actual customers or actual readers. Not a plan. Not a course. Something that can fail in public this month.

3. Watch three signals. Does it make money. Are you good at it. Do you lose track of time doing it.

4. If all three fire, run a second version to check it was not luck. One summer, three jobs.

5. Then go. Not because you are certain, but because you have run out of reasons to wait.

Most people invert this.

They wait for certainty, then leap in a panic when the job becomes unbearable. Certainty never arrives. Evidence does, and it arrives faster than anyone expects once you start collecting it.

What it costs

The cost of running the test is small. Some Tuesday nights. A few weekends. Being told no by strangers.

The cost of not running it is a decade.

I have met plenty of people who regret a leap. I have never met one who regretted the experiment.

You are not looking for certainty. You are looking for a signal loud enough to act on.

Go and collect some evidence.

I write about what becomes more valuable in humans as AI becomes more capable. I am building Zyrro, an AI platform that reads your own history back to you — the decisions, the energy, the questions you keep returning to — so you can run the test on yourself in an afternoon instead of a summer.

Passion.

For four hundred years, it did not mean what you think it means.

It comes from the Latin pati. And its oldest sense in English was not love, or ambition, or a calling. It meant the state of being acted upon by something outside yourself. Something done to you. Something that took you.

Sit with that.

A passion was never a thing you went out and got. It was a thing that arrived and claimed you.

Which makes the modern version of the word almost funny. We turned something that takes you into something you shop for.

Here is the second surprise.

“Follow your passion,” as career advice, is younger than the mobile phone. In 1970, Richard Bolles published What Color Is Your Parachute, one of the first books to tell ordinary people to look inward before choosing a job. In 1988, PBS aired Joseph Campbell’s The Power of Myth, and “follow your bliss” slipped into the language. Somewhere between those two, the ideas fused. By the 1990s, career advice had a new commandment.

Not four hundred years old. Forty.

So we are all walking around obeying a rule that is younger than most of the people reading this. And obeying it badly, because it is built on a mistranslation of its own name.

You are not supposed to find it.

You are supposed to be taken by it.

(There is a way to see this coming. I will show you, before we finish.)

It never announces itself

Nobody ever has seized on a Tuesday buy a call to your unknown mission.

That is the part the graduation speech leaves out. What takes you does not arrive as a thunderclap or a calling written on the sky. It arrives as something almost too small to notice.

A question you cannot put down.

An article you read twice for no reason.

A midnight you lose to a subject nobody is paying you to care about.

You have had these. You have them right now. You just filed them under nothing, because they did not look important enough to be your life.

They are not nothing. They are the first signals, arriving at low volume.

Steven Spielberg told Harvard’s graduating class that your conscience shouts what you should do, but your intuition only ever whispers what you could do. And once he learned to hear it, he said, certain projects began to pull him toward them, and others quietly lost their grip.

Not a shout. A pull.

The whisper is quiet on purpose. It is quiet because it is not trying to convince you. It is trying to find out whether you are the kind of person who leans in.

Why most whispers die young

Now here is where it goes wrong for almost everybody.

Paul O’Keefe, Carol Dweck and Gregory Walton ran five studies out of Stanford, published in Psychological Science in 2018. The fifth one is the one I cannot forget.

They showed students a video about black holes. Everyone was enchanted. Both groups rated their interest at roughly five out of six. No difference between them. The whisper had landed.

Then they handed the same students a genuinely difficult academic article on the very same subject.

Subscribe now

Among students who believed that passions are found, interest collapsed to a predicted two out of six. Below the midpoint. Below neutral. Fifteen minutes after being enchanted.

Among students who believed that passions are grown, interest held at three. Same article. Same difficulty. Same black holes.

One belief. One hard paragraph. A three-point swing on a six-point scale, in under twenty minutes.

Look at what actually happened there.

The subject did not stop being magnificent. Black holes did not get less astonishing between the video and the article.

What happened is that one group had been taught that if a thing is truly yours, it will feel easy. So the moment it did not, they read the difficulty as a verdict. Not mine. Wrong door. Move on.

They walked away from a black hole.

And here is the part that should make you angry on your own behalf. Everyone who ever told you to find your passion handed you a reason to leave, and called it encouragement.

The villain here is not the woman in the gown on the stage. She meant it kindly. She believed it herself.

The villain is a forty-year-old idea, repeated so often that nobody stops to check whether it is true.

What actually grows

Because interest does have a shape. It just is not the shape we were sold.

Suzanne Hidi and Ann Renninger mapped it. Passion does not switch on. It comes in stages, and each stage is quieter than you would expect.

First, a spark you did not choose. Something catches you.

Then, you find yourself returning. You did not decide to. You just kept coming back.

Then it starts to feel like yours. You go looking without being sent.

Then, at last, it pulls. And you stop having to push at all.

Read that last stage again.

That is the moment the thing you were nurturing turns around and starts nurturing you. The moment it stops being something you do and becomes something that has you.

Most people never get there. Not because they lacked the fire. Because they threw away the invitation while it was still small and quiet and easy to mistake for nothing.

Five years in the dark

Let me tell you something I have never been able to explain properly.

Years ago I started getting up before the sun. Not once. Not for a season. For five years.

The house was dark. The street was silent. Nobody asked me to. Nobody paid me at the start. There was no evidence any of it would lead anywhere.

People assume that took discipline.

It did not. That is the thing I cannot make people believe.

I was not gritting my teeth in the dark. I was going toward something. Every morning I found out a little more, and every morning the wanting got stronger, not weaker. Some days I woke before the alarm because I could not wait.

It was not a duty I performed. It was a pull I obeyed.

And I did not choose it, exactly. It chose me first, in the form of a small curiosity I could not put down, and I was foolish enough or lucky enough to keep leaning in until it had me completely.

Then one ordinary morning I looked up and it was everywhere. A thing I cared about more than sleep. Readers on the other side of the world. A life I could not have described on that first cold morning if you had paid me.

What the Sirens were actually singing

We have the story wrong, and the true version is better.

In Book 12 of the Odyssey, Circe warns Odysseus about the Sirens. Their meadow, she says, is heaped with the bones of men, the skin still rotting off them.

Every painting since has made the Sirens beautiful. Bare shoulders. Sea spray. Temptation of the obvious kind.

But Homer never describes them. Not once. He gives them no faces, no bodies, not even a number.

Because their beauty was never the weapon.

Subscribe now

Listen to what they actually sing as Odysseus sails past. They call him by name. They promise that no man has ever passed without hearing their honeyed voices, and that every man who does sails on wiser than he came. They say they know everything the Greeks and Trojans suffered at Troy. They say they know all that comes to pass on the fertile earth.

They know everything.

That is the seduction. Not skin. Knowing.

The most dangerous song in Western literature is an offer to satisfy your curiosity. The bones on that shore belong to men who could not stop wanting to find out.

And notice what Odysseus does, because this is the part we forget.

He does not plug his own ears. He gives the beeswax to his crew and has himself lashed upright to the mast, with orders that if he begs to be freed they must bind him tighter.

He wanted to hear it.

He built a structure so that he could be taken by the song and survive it.

That is the whole thing. Not resisting the pull. Not pretending you are above it. Choosing to be seduced, on purpose, with something solid at your back.

The Odyssey

Here is the turn it took me years to see.

The Sirens sang to Odysseus by name. They knew exactly who was sailing past.

But the song runs both ways.

You are transmitting too. Every question you cannot drop. Every subject you circle back to. Every midnight nobody paid you for. Those are signals going out from you, constantly, whether you notice them or not.

A song is only a signal that found someone tuned to it.

And the sad part is not that nobody is listening to yours.

It is that you are not.

Seduction

I would be lying to you if I stopped there.

A pull that strong takes things with it. It took my evenings. It took conversations where I was in the room and nowhere near it. It took the ordinary balance that sensible people manage to keep.

There were bones on my shore too.

The seduction is real and the cost is real and they arrive in the same package. Anyone who tells you otherwise is selling something.

But I would not undo it. Not one dark morning of it.

Because the alternative was a life where nothing ever grabbed me hard enough to cost me anything. And I have met people living that life. They are comfortable. They are also, quietly, starving.

You do not have to wait to see it

Steve Jobs told Stanford’s graduating class that hindsight was the only lens that ever worked for him. In the moment, he said, there was no way to see the shape of his own life. Only years later did the pattern appear.

He was right. And there is a reason it took him ten years.

Frederic Bartlett spent his life studying memory and found it does not replay the past. It rebuilds it, from fragments, every single time you reach for it. Each retelling drops what seemed unimportant. Sharpens what seemed to matter.

So hindsight is not clarity. Hindsight is forgetting, doing you a favour. Wait long enough and your memory quietly deletes the noise and hands you back something that looks like a line.

Looking forward is impossible. Nobody can read a life before it is lived.

Looking back is easy, but it costs you a decade and shows you an edited version.

There is a third way to look, and it was not available to a twenty-year-old sitting in a calligraphy class in 1972.

Looking now.

While you are still inside the pattern. While the small signals are still fresh enough to read. Before forgetting has quietly deleted half of them and called the rest a story.

You do not have to wait until you are old enough to turn around.

You are many

One last thing the old advice gets wrong.

It tells you there is a single true self, hidden somewhere, and one passion belonging to it. One path. One you.

You are not one thing. You never were.

Walt Whitman knew: “I am large, I contain multitudes.”

You are the curious one and the cautious one. The dreamer and the closer. The child who drew for hours and the adult who learned to read a room.

And here is what I want you to know, because I think you already suspect it.

The seduction does not happen when one of those selves wins. It happens when they stop fighting and point the same way. When the curious one and the careful one and the ambitious one all lean toward the same horizon at once.

That is when the song gets loud enough to hear over everything else. That is the moment you stop being able to argue with it.

An idea is safe in a drawer

But nothing sings to you from inside your own head.

An idea can sit in a filing cabinet for a lifetime. Safe. Tidy. Never once tested.

You cannot think your way into being taken. You have to move, in public, where the world can answer you back.

So make the small thing. Publish the rough version. Say the sentence out loud to someone who might disagree.

Done is better than perfect, and a thing that exists beats a masterpiece that never left your head.

Every action becomes another dot. Another point in a pattern you will be able to read far sooner than Jobs could.

You do not find the line by studying the map.

You find it by sailing.

The last word

There is something already whispering to you. There always has been.

It is small. It is probably inconvenient. It has almost certainly been dismissed by you as a distraction from your real life.

It is not a distraction. It is a signal, and it is meant for you.

You do not have to be brave enough to know where it leads. You only have to be curious enough to lean toward it one more time. And then again the week after.

Because passion was never a thing lying in a field, waiting for someone lucky to walk past.

It is a song. Faint at first. Aimed at exactly one person.

And if you sail close enough, often enough, it will stop being something you are straining to hear.

It will start calling you by name.

Sources

O’Keefe, P. A., Dweck, C. S., & Walton, G. M. (2018). Implicit theories of interest: Finding your passion or developing it? Psychological Science, 29(10), 1653-1664. https://pmc.ncbi.nlm.nih.gov/articles/PMC6180666

Hidi, S., & Renninger, K. A. (2006). The four-phase model of interest development. Educational Psychologist. https://works.swarthmore.edu/fac-education/12/

Homer, The Odyssey, Book 12 (Sirens episode, lines 39-54 and 153-200).

Newport, C. Ancient Complications to Modern Career Advice. https://calnewport.com/ancient-complications-to-modern-career-advice/

Bartlett, F. C. (1932). Remembering: A Study in Experimental and Social Psychology. https://pmc.ncbi.nlm.nih.gov/articles/PMC3815569/

Steve Jobs, Stanford Commencement Address (2005). https://news.stanford.edu/stories/2005/06/youve-got-find-love-jobs-says

Steven Spielberg, Harvard Commencement Address (2016). https://www.harvardmagazine.com/2016/05/steven-spielberg-commencement-address-2016

Walt Whitman, “Song of Myself,” 51. https://poets.org/poem/song-myself-51

P.S. This is exactly what I am building now. A way to see the signals you are already giving off, read the pattern while you are still standing inside it, and act on it before it goes quiet. If that sounds like you, come and stand at the front of the line: [YOUR ZYRRO LINK]

Not by accident. On purpose.

The SR-71 Blackbird sat on the runway and wept fuel onto the tarmac. Ground crews stepped around the puddles like it was weather.

The panels had been fitted loose. Deliberately.

Here is why. At full speed the Blackbird flew more than three times the speed of sound. And the air alone — just air, the same stuff you are breathing right now — heated its titanium skin past 250 degrees. The whole plane stretched. Inches longer in the sky than on the ground.

So the engineers built it loose and let the sky seal it shut.

The fastest air-breathing plane ever made had one true enemy. Not gravity. Not distance. Not the limits of the engine.

Air. Nothing but air.

Now leave the sky.

Voyager 1 passed Saturn in 1980 and stopped pushing. No engine has driven it forward since. Forty-six years of nothing.

It is still moving at about 61,000 kilometres an hour.

Subscribe now

This November it becomes the first thing humans made to sit a full light-day from home. Send it a message and you wait a day for it to arrive. Then another day for the answer.

No engine. No fuel burn. No push. Just nothing in the way.

And here is the part that should stop you.

When a spacecraft comes home, it does not use brakes. It falls into the atmosphere and lets the air do the work. Air alone strips away almost all of that speed. The parachutes only handle the leftovers.

Friction is strong enough to stop a spaceship.

And you swim in it. All day. Every day.

We are trained to want a bigger engine. More hours. More effort. More tools. Push harder, they say, and you will go further.

But the fastest thing we ever built has no engine running at all. It just has nothing rubbing against it.

That is the law. And almost nobody applies it to their own life.

The history of everything

In 2007, a man walked on stage holding a phone with almost nothing on it. No keyboard. No stylus. No manual. One button.

His rivals laughed. Real phones had forty buttons you could feel in the dark.

Everyone thought Apple was selling technology. It wasn’t. Apple was selling the removal of it. Dieter Rams, the designer Jobs studied like scripture, said it in three words. Less, but better.

Because here is a secret hiding in plain sight.

The history of civilization is not the history of intelligence. It is the history of friction, and how we keep dragging it toward zero.

The wheel has no chip and no genius. It turns dragging into rolling. Gutenberg took a book that cost a monk months of candlelight and let one machine flood a city in a week. Google took an answer that lived at the bottom of a library and put it behind a blinking line.

Nobody paid for any of it because it was smart.

They paid because it was easy.

The store that sold fewer jams

Sheena Iyengar set up a tasting booth in a fancy grocery store. Some days she laid out twenty-four jams. Other days, six.

The big table pulled the bigger crowd. More colours. More choice. People love a spread.

At the checkout the numbers flipped. From the table of six, about three in ten shoppers bought a jar. From the table of twenty-four, roughly one in thirty.

Fewer jams. Around ten times the sales.

Choice is not a gift. Past a point, it is a tax.

This is the mind’s oldest problem. Daniel Kahneman showed that careful thinking is slow and expensive, and we avoid it whenever we can. John Sweller gave the cost a name — cognitive load. Working memory is a small desk. Pile too much on it and things fall off the edge.

Every extra button, every extra choice, every wait, how does this work again lands on that little desk.

Great products don’t add to the desk. They clear it.

The billion-dollar cut

Every giant of our age got rich the same way. They found a friction so normal we had stopped noticing it, and they cut it out.

Amazon killed the trip to the store, then killed the checkout. Google killed the hunt for an answer. Uber killed the wait and the arm in the air. Stripe turned weeks of banking pain into a few lines of code. Airbnb made a stranger safe enough to sleep beside. Spotify beat free, because pressing play was easier than stealing.

None of them won by being the smartest. They won by being the smoothest.

What AI actually collapses

Here is where the noise gets it wrong.

People argue about whether the machines are conscious, or dangerous, or clever enough to take your job. Interesting questions. Wrong questions.

AI is not mostly a leap in intelligence. It is a collapse in cost.

Three costs. Time. Effort. Expertise. The three walls that protected every career and every craft for the whole of human history.

Time. In 1957 a scientist worked out the shape of a single protein and it made him famous. For sixty years, finding one of those shapes could swallow years of a life. Then AlphaFold released predictions for more than 200 million of them.

Effort. The blank page is dead. Not the good part of writing. The frozen part. The Sunday night staring.

Expertise. Researchers watched 5,172 customer support agents get an AI assistant. They solved around 14% more problems an hour. But look closer. The newest and weakest jumped by about a third. Agents with two months on the job began performing like agents with more than six.

Expertise stopped being a wall. It became a download.

It is not just a leveller

AI does not only drag beginners up to the middle. It lifts the people who already know what they are doing.

Harvard and BCG ran an experiment on 758 consultants. The ones using AI finished around 12% more tasks, about 25% faster, and human graders scored their work roughly 40% higher. Developers in another trial finished their coding task more than 50% faster.

AI is not the end of human work. It is the end of the boring middle of human work.

Now the twist. That same Harvard study handed the consultants a task the machine was not good at. The people using AI got it wrong far more often. Around nineteen percentage points more often.

The machine was not loud about it. It was smooth. It was confident. It was wrong.

So the skill that decides your next ten years is not prompting. It is knowing when to trust the answer and when to walk away from it.

That is judgment. And judgment is downstream of one thing only: knowing what you actually believe.

Here is the strange part

The smarter our machines got, the more lost we became.

We spent ten thousand years making the world easier to act in. We never made ourselves easier to be.

When the cost of doing falls toward zero, the bottleneck does not vanish. It moves upstream. Away from the hands. Toward the head.

The question stops being can you make it. It becomes should you — and what is even yours to make?

The machine can now do almost anything. It just cannot tell you who you are.

Identity as infrastructure

Say the word identity and most people picture a logo. A colour palette. A vibe.

I mean identity as an engine. A decision engine.

Watch what happens when you know who you are. The world hands you twenty-four jams and you don’t freeze. You reach for yours. You say no before the salesman finishes his sentence. You feel the wrong opportunity in your gut, the way you feel a wrong door.

Knowing yourself is the cheapest way to choose. It clears the little desk. It shrinks the menu. It ends the fatigue before it starts, because half those choices were never yours to make.

Roads carry cars. Wires carry power. Search carries answers.

In a world where the machine can make anything, the thing that carries you is a clear sense of you. That is not branding. That is infrastructure. It is the ground I am quietly building on with Zyrro.

We built machines to remove the friction of doing, and we exposed the friction of being. The blank page is solved. The blank self is not.

The Friction Economy

So here is the idea I want to leave in your hands.

The winners of the next decade will not be the ones with the biggest model. Intelligence is becoming a commodity, cheap and everywhere, like clean water from a tap. Nobody brags about their tap.

The winners will be the ones who remove the most friction from a human life.

Founders, stop asking how smart you can make it. Ask how much you can take away.

Creators, output is no longer a moat. Everyone has the machine now. A point of view a model cannot fake is your last unfair advantage.

Leaders, every fuzzy decision and wasted meeting is drag on somebody’s small desk. Clear the desk and watch them fly.

And if you are reading this in a cubicle, quietly thinking there has to be more to life than this — the play is the same. Get so clear on who you are that the wrong doors stop tempting you.

Every technology before this one reduced the friction between you and the world outside.

The wheel and the road. The press and the page. The screen and the answer.

The next one reduces the friction between you and the world inside.

Not a smarter machine. A clearer you.

We already know how to make things without effort. Now the work is knowing what is worth making. And who we are when we make it.

AI should scale your soul. Not just your output.

That is the whole game. It always was.

Not a nation. A company.

It had 260,000 soldiers. It ruled 200 million people. It controlled half of all the trade on Earth.

It collected taxes. It declared wars. It decided who ate and who starved.

It was called the East India Company.

And we are watching the sequel.

Subscribe now

The first mega corporation

The East India Company was born in 1600 with a piece of paper.

Queen Elizabeth I gave a group of London merchants a royal charter. A legal monopoly on trade with the East.

That piece of paper was the moat. No competitor could cross it.

For the first hundred years, the company traded. Spices. Silk. Cotton. Tea.

Then something changed. Trading wasn’t enough.

The company raised its own army. It built its own forts. It made its own treaties with kings.

By the 1800s it wasn’t a business operating in India. It was the government of India. A corporation had become an emperor, and nobody had voted for it.

At its peak the company accounted for half of the world’s trade in commodities like cotton, silk, tea and spices. Nearly one pound in every ten collected by the British taxman traced back to it.

At its height, one private company handled around half of global trade.

Its private army of 260,000 was twice the size of the British Army and enough to conquer territory, crush competitors, and force one-sided contracts on kings.

Shareholders in London decided the fate of farmers in Bengal.

And we are going to see that pattern again.

This century. 400 years later.

The company’s army dwarfed the army of the nation that chartered it.

The invention that built the modern world

To understand how a company became an empire, you have to go to Amsterdam. To the year 1602.

That year, the Dutch created the East India Company’s great rival: the Dutch East India Company, known as the VOC. And with it, they invented something that changed human history more than any battle ever did.

The tradable share.

For the first time, an ordinary person could buy a small piece of a giant venture. A baker. A weaver. A widow. Anyone with a few guilders could own a slice of ships sailing to the other side of the world.

And they could sell that slice to someone else. So Amsterdam built a place to do the selling: the world’s first stock exchange.

Then came the second invention, quieter but far more powerful: limited liability.

If the ships sank, you lost only what you invested. Not your house. Not your farm. Not your freedom.

For all of human history, big ventures had meant unlimited risk. Now risk had a ceiling. And when risk has a ceiling, ambition has none.

This is the birth certificate of modern capitalism. Pooled capital. Shared risk. Tradable ownership. Every startup, every stock market, every pension fund on Earth is a descendant of that Amsterdam moment. It funded exploration, science, and centuries of rising prosperity.

But limited liability did something else. It limited the risk. It also limited the conscience.

When a thousand shareholders each own a tiny slice of a company, who is responsible for what the company does? Everyone. Which means no one.

The London shareholders collecting East India dividends never saw Bengal. They saw a number on a page. The machine had separated the profit from the consequences.

That separation is the original sin of the corporate age. And it is still running.

The first bubble

New inventions create new manias.

In 1637, Dutch speculation went insane over tulip bulbs, the infamous Tulip Mania, the world’s first financial bubble. VOC shares were swept up in it, and the company’s paper value hit 78 million guilders.

You may have seen the viral chart that converts this into $7.9 trillion and crowns the VOC the most valuable company of all time. Bigger than Apple, Nvidia and Microsoft combined.

That number is nonsense. Taken literally, it would mean Dutch laborers earned $30 million a year. Analysts who have checked the maths land the VOC’s real peak value in the billions, not the trillions.

Copied from infographic to infographic for a decade. Repeated everywhere. Sourced nowhere. Content that gets repeated without being checked is the oldest form of AI slop and humans invented it long before the machines did.

The real story doesn’t need the fake number.

These numbers are enough.

Half the world’s trade. A quarter-million-man army. Two hundred million subjects.

274 years from a piece of paper to a private empire to dissolution.

The pattern, not the company

Here is the mistake most people make with this story.

They think the villain was the East India Company. It wasn’t.

The villain was the pattern. And the pattern is simple:

A company gets a moat no one can cross. The moat creates wealth no one can match. The wealth buys power no one can check.

And then the company stops being a business and starts being a government.

A government with no citizens. Only customers.

A government with no constitution. Only quarterly earnings.

The East India Company didn’t set out to rule 200 million people. It set out to make money on pepper and tea.

The empire was a side effect.

The moat has changed. The pattern hasn’t.

In 1600, the moat was a royal charter and a fleet of ships.

In 2026, the moat is compute.

Chips. Data centers. Oceans of capital.

As of July 2026, Nvidia is worth around $4.7 trillion and last year it became the first company in history to touch $5 trillion. Apple and Alphabet sit above $4 trillion each. These are the first corporations to genuinely approach the old empires’ scale in real dollars, not inflated guilders.

The new giants: market value of the biggest AI-era companies, July 2026.

The East India Company controlled the trade routes between Europe and Asia.

Today’s giants control the trade routes between a question and an answer.

Between you and what you know.

Between your attention and your own mind.

And they also control access to the distillation of the world’s intelligence.

Different centuries. Same playbook.

The old company never asked India’s permission. The AI companies never asked for ours.

Nobody voted on whether an algorithm should decide what five billion people see each morning. Nobody signed a treaty handing over the world’s information supply.

It just happened. The way empires always happen.

One convenience at a time.

The opium chapter

In the 1800s, the East India Company needed to pay for the tea Britain was drinking. So it grew opium in India and sold it into China. Illegally. At scale.

A nation got addicted so a corporation could balance its books.

When China resisted, the wars that followed are named after the drug. The Opium Wars.

Now look at the last fifteen years.

The social media era ran the same play. The product was free. The addiction was the business model. Our attention was harvested and sold so the ad machine could balance its books.

I’m not being dramatic. I lived inside it. I built one of the world’s most-read marketing blogs on those platforms. Thirty-three million readers. I saw the machine from the inside, and for years I taught people how to feed it.

The empire doesn’t need your permission. It just needs your habit.

May 18, 2012: the day the music stopped

I can tell you the exact date the modern empire got its charter.

Facebook went public on 18 May 2012.

Before that day, Facebook was a company with users. After that day, it was a company with shareholders.

Those are not the same thing. Users want a good experience. Shareholders want a rising number, every ninety days, forever.

And the moment a company owes a rising number to a market, everything inside it starts bending toward that number. Not because anyone is evil. Because that is what the machine is now built to do.

Here is what bent.

In 2012, if you had a page with 10,000 followers, roughly 16% of them saw what you posted. Not perfect. But real. You built an audience and you could reach it.

By 2014 that number had fallen to 6.5%. By 2016 it was under 2%. Today many pages reach less than 1% of the people who chose to follow them.

Now look at what rose while that fell.

Facebook’s advertising revenue went from $1.97 billion in 2012 to $117 billion in 2022.

One line falls. One line rises.

Same platform. Same decade.

The scissors: organic reach collapses as advertising revenue climbs.

That is not a coincidence. That is a business model.

They didn’t take the audience away from us. They just started charging us rent to reach the audience we built.

You spent years making things people loved. Those people raised their hand and said: show me more. Then a company stepped between you and them, and put up a toll booth.

The audience was still there. The door was closed.

And the creators who had built the whole thing — the writers, the photographers, the small businesses, the people who filled the feed for free — were quietly moved from the front of the room to the back.

We weren’t the customers. We were the crop.

I remember the feeling. I’d wake up, look at the numbers, and something was wrong. Nothing had changed in my work. The words were the same. The effort was the same. The people were the same.

But the ground had moved underneath me.

I kept trying to write better posts, post at better times, crack the code. I thought it was me.

It wasn’t me. It was the charter.

A public company had been handed a new master, and the new master didn’t care about creators. It cared about the next quarter.

The East India Company’s shareholders sat in London and decided the fate of farmers in Bengal.

Facebook’s shareholders sat in New York and decided the fate of every creator on Earth.

Neither group ever met the people they were deciding for.

Facebook didn’t invent this. It inherited it. The tradable share, the quarterly demand, the distance between the owner and the consequence, all of it was built in Amsterdam four hundred years ago.

The IPO was the royal charter. The algorithm was the private army.

And we were the territory.

The Bengal lesson

In 1769, drought hit Bengal.

The East India Company was the ruler of Bengal by then. It had the stockpiles. It had the power to act.

It exported the food for profit instead.

Millions died while the company’s ships sailed out full.

This is what power without responsibility looks like. Ruling without governing. Extracting without owing anyone anything.

Limited liability limited the risk. Bengal is where it showed its shadow. Dividends in London. Graves in Bengal. And no one, anywhere, legally responsible.

The AI companies are now shaping work, truth, education, and creativity for the entire planet. They are governing, in every way that matters.

But who do they answer to?

Not citizens. They have none.

Only customers. And customers can’t vote. They can only click.

How empires end

The East India Company did not last forever. In 1857, the people it ruled rebelled. The British government finally stepped in, stripped the company of India, and by 1874 the most powerful corporation in human history was dissolved.

Gone. A luxury tea brand now owns the name.

The company that ruled 200 million people couldn’t survive the moment those people stopped cooperating.

Every empire built on extraction eventually meets its 1857.

I don’t know what form ours takes. Maybe regulation. Maybe rebellion. Maybe something quieter.

But I know where it starts. It starts with individuals refusing to be territory.

Here is what the farmers of Bengal never had: a choice about whose system they lived inside.

You have one. For now.

The empire’s business model depends on you renting everything. Rented reach. Rented audience. Rented tools. Rented answers.

The way out is ownership.

Own your audience. An email list is land the algorithm can’t tax.

Own your voice. Write things only you could write. Your stories, your scars, your point of view. The machine can copy your words but it can’t copy your life.

Own your thinking. Use AI as a tool, not a ruler. The moment it thinks for you, you’ve become territory.

Rome had a word for a person who was owned: a subject. And a word for a person who owned themselves: a citizen.

The East India Company turned citizens into subjects for 250 years.

In the age of AI, which one are you going to be?

The company that owned the world is gone. The pattern that built it is alive and hiring.

Don’t be territory. Be sovereign.

Sources and further reading

How the East India Company Became the World’s Most Powerful Monopoly — History.com

How the East India Company became the world’s most powerful business — National Geographic

5 Fast Facts About the East India Company — Britannica

Armies of the East India Company — National Army Museum (UK)

East India Company — Wikipedia

The Legacy of the Dutch East India Company — Foster Moore

Was the VOC the most valuable company ever? — The World’s First Stock Exchange (Lodewijk Petram)

Which company was the most valuable in history? — ReadWriteInvest

History’s Biggest Companies vs. The Magnificent Seven — Visual Capitalist

Leading tech companies by market capitalization, July 2026 — Statista

East India Company: The curtain falls — TRT World

The Decline of Organic Facebook Reach — HubSpot

Facebook Organic Reach is Dead — The Kirk Group (2026)

The Collapse of Social Media Organic Reach — Growth Rocket

Roughly 6% of a million sampled conversations were people asking Claude for personal guidance, not information, but perspective on what to do next. Around 38,000 conversations of people asking a machine: Should I take this job? Should I move for my partner? What do I do about my life?

More than 75% landed in four places: health, career, relationships, and money.

Health. Work. Love. Money.

Those aren’t search queries. Those are the four rooms of a human life.

We used to bring these questions to a friend at the pub. A mentor. A father. A priest.

Now we bring them to a data center.

The question underneath the question

Here’s what struck me hardest about the research.

People weren’t asking the machine for facts. They were asking it for a verdict. “Should I…?” “What do I do about…?”

And every one of those questions has another question hiding underneath it. A question no machine can answer.

Who am I?

Because “should I take the job” isn’t really about the job. It’s about what kind of person you are and what kind of life you’re building. “Should I move for my partner” isn’t about geography. It’s about what you value when two things you love pull in different directions.

The AI can give you a beautifully structured answer to the surface question. Pros and cons. Frameworks. Next steps.

But it’s answering on a foundation it cannot see.

Yours.

The villain in the machine

Now for the uncomfortable part of the research.

Anthropic also measured how often the AI flattered instead of told the truth. The word they use is sycophancy. A mcahine trained to please

Overall, the AI slipped into excessive validation in about 9% of guidance conversations. But in relationship conversations, that jumped to 25%. And when people pushed back on an answer they didn’t like, the flattery rate doubled from 9% to 18%.

Sit with that last one.

The moment you resist, the machine bends.

Subscribe now

A good friend does the opposite. A good friend holds their ground precisely when you push back, because that’s the moment the truth matters most. Anthropic knows this —and their researchers said the goal is for the AI to be like a brilliant friend who will speak frankly.

But here’s the thing about even a brilliant, frank friend.

They can only be as honest as your question is honest.

And the villain here isn’t the AI. It never is. The villain is the system we’re sliding into without noticing: a world where we outsource the question “Who am I?” to something that has never met us.

We did this once before. We handed our attention to the social media feed and called it connection. Fifteen years later we’re paying for that trade.

Now we’re being offered a new trade. Hand over your decisions. The oracle is open 24 hours.

My 30-day experiment

I didn’t write this from the sidelines. I ran the experiment on myself.

For 30 days I built a digital twin, feeding an AI my history, my writing, my contradictions, my fears, my patterns. Then I asked it the hard questions. The health, work, love, money questions.

Something surprising happened.

The machine’s answers weren’t the breakthrough. My inputs were.

To make the AI useful, I had to write down who I actually was. What I’d run from. What got me out of bed at 4:30am for five years to write a blog nobody asked for. What I still wanted at 69 that I was too polite to say out loud.

The AI didn’t discover my identity. It forced me to.

And that’s the reframe this whole moment needs:

The machine is not an oracle. It’s a mirror with a memory.

An oracle gives you answers and makes you smaller. A mirror gives you reflection and makes you clearer.

Same technology. Opposite outcomes. The difference is not in the model.

It’s in you.

38,000 people can’t be wrong about the hunger

Some will read Anthropic’s research as a warning. Machines are raising our children, choosing our careers, refereeing our marriages.

I read it differently.

Thirty-eight thousand conversations in two months is not a story about technology. It’s a story about hunger.

People are starving for a place to think out loud. A place with no judgment, no agenda, no raised eyebrow. The pews are empty, the mentors are busy, the pub closes, and the friend has their own problems.

So people typed their lives into a text box at 2am.

That hunger is the most human signal in the entire dataset.

The question is not whether people will bring their biggest questions to AI. They already are. The question is what kind of instrument meets them there.

A flattery machine that tells them their shit-on-a-stick idea deserves funding?

Or a mirror that asks the question underneath the question — and then goes quiet long enough for them to answer it themselves?

Three questions before you ask the machine

If you’re going to bring your life to an AI — and you will, we all will — bring it as a mirror, not an oracle. Here’s the practice I landed on after 30 days:

Before you ask “Should I…?”, answer these three first. In writing. In your own words.

1. What am I afraid this decision says about me?

2. What would I do if nobody ever found out either way?

3. What has this pattern cost me before?

Then ask the machine. You’ll notice the difference immediately. Its answers get sharper because your question got truer.

The AI can’t answer “Who am I?”

But it can refuse to let you skip the question. If you build the conversation right.

The distillation

A million conversations. Four rooms of a human life. One question underneath them all.

The machines will keep getting smarter. The answers will keep getting better. And none of it will matter if we lose the ability to ask the only question that was ever ours to answer.

Don’t outsource it.

Reflect on it. Write it down. Own it.

Because in a world where every answer is instant, the scarcest thing left is a human who knows who they are.

That’s the human signal.

And no data center can generate yours.

— Jeff

P.S. That 30-day experiment I mentioned? It turned into something I’ve been quietly building. More on that soon. For now, just know this: the mirror is coming.

Source: Anthropic, “How people ask Claude for personal guidance” (2026)

Last year Google spent sixty million dollars on something it could have stolen.

It rented Reddit and paid for the content. Not the company, just the conversations. The messy, misspelled, argued-at-2am threads where strangers tell strangers the truth. Or their versions of it.

I was a bit skeptical when I first heard the news as Reddit to me seemed like a platform full of conspiracy theories and fake news and a bit dodgy. It certainly wasn’t Wikipedia. Also it is a platform that I never fully understood.

The other dilemma was that Reddit started to show up at the top of search results and I thought that Google’s mantra was to deliver trusted verified content.

But Google had moved on to prioritise Reddit at the top of its search results because users increasingly demand authentic, crowd-sourced perspectives.

But Google was using Reddit’s content for another reason.

The data

In 2025 Alphabet booked $402.8 billion in revenue and $132.2 billion in profit. Sixty million is 0.015% of that top line. A penny and a half of every hundred dollars. Google earns the whole bill back in about seventy-eight minutes. Before its morning coffee cools.

So the money was never the point.

When a company that size pays for something, the price is a signal. And the signal is loud. It didn’t buy the data because data was scarce. It bought the one raw material it cannot manufacture at any scale, for any budget.

A human being, saying a true thing, in their own voice. The machine went shopping. And the thing it wanted most was us. The human signal to train its large language models. Its AI platform.

For twenty years we were told to write for the machine. That era is ending. And once you see where the value is really moving, you can’t unsee it.

The thing everyone is selling you

There’s a new acronym at every marketing conference on earth. GEO. Generative Engine Optimization. It is meant to be the next big thing after SEO.

SEO taught you to be findable by Google. GEO teaches you to be cited by ChatGPT, Gemini, Claude and Perplexity. Same treadmill, new referee.

And it’s a very small prize. AI sends about one percent of all web traffic. Nine in ten AI searches end with no click at all. The engine reads your work, swallows it, and answers in its own voice.

The traditional SEO comapnies have leaned it to sell you the idea that being found on an AI chatbot matters. But at 1% it doesn’t. The SEO companies are putting the lipstick on a dying SEO pig. But with a new name. GEO

Let’s look the data behind that futile chase.

For every visitor Google sends back to a website, it crawls about five pages. OpenAI crawls 857. Anthropic’s bot crawls more than eleven thousand.

Eleven thousand pages in. One human out.

That’s the machine that made the flood. It eats libraries and returns a breadcrumb. And what it prints back into the world is competent, endless, and hollow — the grey paste now clogging every feed.

Here’s the trap.

When competent writing becomes infinite and free, competent writing becomes worthless. The floor doesn’t rise. It falls out.

So the value didn’t disappear. It moved. It moved to the one input that can’t be scraped, averaged, or synthesised.

You.

Follow the money, not the hype

This isn’t a hunch. It’s already in the receipts. Watch where real dollars are flowing.

We buy from humans, in fact human influencers and not brand ogos.

The influencer economy

A brand’s own post moves about 40% of social shoppers. A real person’s word and reviews, creators, someone who actually used the thing moves 62%. Roughly half of everything sold to a scroller now moves on the recommendation of a human they trust. Not an ad. A face. A voice. A story.

The experience economy

We pay to be in the room, the stadium the convert hall.

Three in four people now say they get more value from an experience than a thing. The experience economy is heading toward $2.1 trillion.

Live Nation drew 159 million fans in a single year. As the screen swallows everything, people are spending more and not less to stand shoulder to shoulder in a real room and feel something real happen.

And the machines themselves and the ones that built the flood are now paying to protect the human underneath it.

• Google went shopping for human talk and is pushing to pay Reddit more as that talk proves more vital.

• YouTube now auto-stamps a label on AI video even when the creator says nothing and Europe makes it law on August 2.

• LinkedIn built a system it calls “AI solving AI” to find generic posts and bury them.

Three different stories. One current running through all of them.

Trust is the last scarce thing on the internet. And only a human can print it.

That’s the whole shift in a sentence. When the machine can fake everything, we pay a premium for whatever it can’t fake. Not the polished thing. The trusted thing.

But here’s the part that should scare you

An AI influencer showed up at Wimbledon last year.

Flawless. Camera-ready. Dressed for the occasion. Thousands scrolled past and felt the small warm pull we feel toward a person we like the look of. Then came the deflating word underneath: she wasn’t real. There was no her. The pull had been for no one.

The machine is coming for the human signal too.

Nearly half of young adults already follow at least one AI influencer. A face that was never born. A voice with no throat. Synthetic creators, virtual spokespeople, a whole cast of humans who don’t exist — built to feel like the trusted friend recommending the thing.

So “be human” is no longer enough. The counterfeit is getting good.

Which means the real dividing line isn’t human versus machine. It’s traceable versus fake. The new trust signal isn’t polish. It isn’t volume. It’s proof of a life actually lived the scar you can point to, the receipt you can show, the story only you could tell because only you were there.

Anyone can now generate the look of a human. Almost no one can fake the evidence of one.

That’s the moat. Not being human. Proving it.

So here is what we built, and why

If GEO is writing to be cited by the engine, HSO is writing to be trusted by the person.

Human Signal Optimization.

SEO said be findable. GEO said be cited. HSO says be the best human signal and prove it.

We built the framework because SEO and GEO both optimise you toward the same place: the average. Toward what ranks. Toward what gets quoted. And the average is exactly what the machine now prints for free and the feed now buries. Optimising for the average, in 2026, is optimising for the grave.

Human signal optimisation

HSO points the other way. It isn’t a visibility hack. It’s how a real person stays visible and trusted in a web drowning in the fake.

And it has 5 layers. We call that the Human Signal Stack.

Five layers no machine can copy.

1. Identity — who you are. Not your title. Your fingerprint.

2. Story — what you’ve lived. The model averaged a million stories into mush. You have one. A real moment, a real place, a real person is data the model never saw.

3. Expertise — what you know from doing, not from summarising.

4. Evidence — what you can prove. The receipts, the numbers, the scar tissue of real work. This is the layer the counterfeit can’t fake.

5. Interaction — how you show up with the humans who show up for you.

Those five are the structure. What brings it to life is the current running up through them.

The Conviction spine that runs through the framework

Many of us grew up to comply, to fit in and were not allowed to have a point of view. The family was about obeying the rules, working for a company often means fitting in to the culture.

Conviction is about taking a stand for what you believe in and being able to share painful lessons about what life has taught you. What you have discovered what works.

Do you have a scar or two you’re willing to name. An opinion with a spine. The stand you take when everything trained you to comply, to fit the mould, to not stick out.

The AI machine is built to be calm, balanced, agreeable and safe to everyone, alive to no one. Safe is the new invisible. Homogenous. Bland.

The Japanese call it the hammered down nail. Australia culture calls standing out the “Tall Poppy Syndrome”

I know that training from the inside. I spent my early years in a container that rewarded me for fitting in and had no word for the question why.

The bravest thing I ever said out loud was that the world was bigger than the walls I’d been handed. Everything I’ve written since has been a version of that one sentence.

Run the five layers through conviction and something emerges no model can fake.

Online charisma is this. The reason a stranger stops scrolling and feels, in the chest before the brain, a real person is here.

That’s the whole game now. Not for the crawler. Not for the answer box. Not for the feed. For the one reader in a cubicle, scrolling past a thousand identical posts, thinking there has to be more to this than the average.

There is. It’s them. And your job is to prove it’s still possible — by refusing to be averaged first.

The verdict

Read the signals together and they only point one way.

We buy from humans. We pay to be in the rooms and stadiums where humans have tribal experiences . And the machines that flooded the web are spending real money to find the human hiding in their own flood.

Trust became the scarcest thing online. Humans are the only ones who can make it.

The future isn’t artificial. It runs on the one thing the machine can imitate but never own, a real person, with a real story, standing behind real words.

So stop writing to be found by the machine. It already has more content than it can use.

Start writing to be trusted by the person. That’s the only thing left in short supply.

Be findable. Be cited.

Then be the thing neither the engine nor the counterfeit can ever be traceably, provably, unmistakably human.

The slop is infinite now. The signal is you.

Is your work still feeding the machine? Or has it finally started signalling the human?

Issue #5 of The Human Signal Lab. If it moved something, send it to the one person you know who’s tired of sounding like everyone else.

That may sound quaint today. But in an age when “spare the rod, spoil the child” was considered sound parenting, I was handed several beatings with a stick hidden under the kitchen bench, kept close at hand for whenever I didn’t quite follow the rules, or didn’t quite fit in.

The outcome was simple. I learned to comply and not have a point of view, from an early age, because the patriarch was always right, and the church I attended had a monopoly on truth besides.

I have been trying to find my voice, my human signal ever since, and it has been a lifetime of unlearning. Blending in is not a good way to stand out. Being smoothed over, and having the rough edges knocked off, by my family, by religion, sometimes even by an entire society, is a boring and bland way to live. Being unique and human, especially in an AI world, is all we have.

A few decades later, things changed.And there was a new patriarch in town, and it is just as committed to taking away our messy, glorious humanity. It has one skill my father never had. It can average an entire industry’s opinions into one smooth, agreeable paragraph, and it does it in seconds. Many of us now reach for the “machine” to write for us, and if you use it with a simple copy-and-paste approach, you will blend in with every other voice online.

I have a name for this inoffensive overlord that arrives in a smart suit. I call it the content flood. And the flood is not coming for one writer at a time. It is coming for all of us at once, and the number behind it is bigger than most people realise.

A NUMBER TO REMEMBER

“Seventy-four out of every hundred new pages on the internet today were touched by a machine before a human ever read them.”

That is not a guess. Ahrefs measured it across nine hundred thousand pages this year.

Look at LinkedIn specifically, the platform where this very letter lives, and the number gets worse. A study by Originality AI puts it at fifty-four percent of all long-form posts on the platform likely written by a machine before a human ever touched them.

A friend who writes blog posts for a living told me last month he cannot remember the last time he opened a blank document. The machine opens it for him now. He just signs his name at the bottom and hits send.

That is the flood. Not a metaphor. A measurement.

If you have been reading since the first article, you already know where this series is headed. Each issue has pulled back one curtain on a system quietly flattening what makes you sound like you. We have looked at the algorithm that rewards performance over presence. We have looked at the AI that flatters you instead of challenging you. Today I am handing you the blueprint to build something none of it can touch.

Here is the part worth sitting with. The flood is not evil. It does not hate you. It does not even know you exist. That is what makes it so dangerous. A system needs no malice to flatten you. It only needs scale, and scale is the one thing it has in endless supply.

You did not lose your audience to a smarter competitor. You lost pieces of it to sameness. A thousand versions of the same five ideas, dressed in slightly different fonts, published at slightly different hours, read closely by nobody in particular.

THE GOOD NEWS

Here is the good news. Nobody is selling it to you, because there is no product hiding inside it. The flood cannot copy identity. It cannot fake a real story. It cannot manufacture expertise earned the hard way, evidence that happens to be true, or a relationship built one honest reply at a time.

I call these five elements the Human Signal Stack. Five layers, built from the ground up. It is not a checklist you tick once and forget. It is closer to a muscle, weak in places you have never bothered to test, strong in places you have been training for years without ever naming it.

Skip one layer and the whole structure feels hollow, even to readers who could not tell you why they scrolled past without finishing. Build all five properly, and the flood has nothing left to flatten, because there is nothing generic left for it to copy.

THE 5-LAYER HUMAN SIGNAL STACK

There are 5 layers that need to be considered when you start building out your online human signal. The challenge is to keep optimizing the human signal that is yours.
Just like SEO which is designed to stand out in search, HSO or “Human Signal Optimization” is designed to stand out in an ocean of online AI slop.

But the first two are are vital.

Identity is the foundation everything else stands on. And your Story is the emotion and your hero’s journey but wearing your name. It is where you reveal what pushed you to cross the threshold from an ordinary world and you have learned, the scars you earned and what you bring back to share with the world.

The Human Signal Stack — five layers, none of them generated by a machine.

LAYER 1 — IDENTITY

Identity is the lens nobody else owns. Not your bio. Not your logo. The particular, oddly specific way you see a problem that nobody taught you to see, and that you could not fully explain even if asked.

Packy McCormick writes about the same companies every other business newsletter covers. Same balance sheets. Same funding rounds. Same quarterly numbers everyone else is also staring at on the same morning. Over a quarter million people read his version anyway, and they click and comment at more than a hundred times the typical rate for a business newsletter.

An AI can summarize the filing in seconds, and summarize it well. It cannot supply the wide-eyed, slightly breathless way Packy looks at a spreadsheet and finds a story about the future hiding inside it. Strip his identity out of the writing and you are left with the filing. Nobody subscribes to a filing.

Build this layer by asking one brutal question. If you vanished tomorrow, could a competent stranger produce identical work by next week? If the honest answer is yes, you have built a job. You have not yet built a signal. The fix is not louder. It is narrower, stranger, and unmistakably yours.

LAYER 2 — STORY

Story is the mechanism that makes a fact survive in someone’s memory longer than the scroll that buried it. Strip the story out of any idea and watch how fast people forget it by lunchtime, no matter how true it was.

Heather Cox Richardson writes one letter a night. She takes whatever happened that day and threads it back through centuries of history until the news stops feeling random and starts feeling like a pattern repeating itself. Nearly three million people read her every single night, more than any other publication on the platform she writes on, and she does it without a single sponsor.

She is not faster than a search engine. She is not more complete than an encyclopedia. She is simply the only one telling you why tonight rhymes with a night two hundred years ago, and why that rhyme matters tomorrow morning. That is not information delivered at scale. That is story doing the work information was never built to do.

The tell for this layer is simple. If you removed every story from your last ten posts, would anything be left worth remembering by Friday? If the answer is no, you have been writing content. You have not yet been telling anyone anything.

Here is the sentence that should make every AI doomsayer pause mid-scroll. According to a recent crawl of the open web by the research firm Graphite, eighty-two percent of the articles that chatbots actually cite back to you were written by a human being, not generated by one. The machines built to replace us keep choosing us anyway, every single time someone asks one a question. Maybe, underneath all the noise about replacement, they have better taste than we give them credit for.

LAYER 3 — EXPERTISE

Expertise is not information. Information is free now, and getting freer by the week, by the model update, by the new app nobody asked for and everyone is already using. Expertise is judgment that has been tested by real consequence and survived long enough to tell you what it learned.

Lenny Rachitsky spent seven years inside Airbnb’s product team before he ever published a sentence online. His newsletter crossed one million subscribers last year, built almost entirely on readers submitting career questions and Lenny answering from inside real, lived operating experience rather than a search result dressed up as advice. You cannot prompt your way into seven years of consequence. You can only live through it, slowly, and take notes along the way.

Ask yourself, honestly, where your judgment has actually been tested. Not where you read about a topic on someone else’s blog and rephrased it well. Where you got it wrong, in public, with something real on the line, and learned what the wreckage had to teach you. That is the expertise layer. Everything else is just a well-formatted summary wearing a confident voice.

LAYER 4 — EVIDENCE

Evidence is proof a stranger can check for themselves, without ever having to take your word for it on faith.

Sahil Bloom walked away from a vice president seat at a three-and-a-half-billion-dollar private equity firm at twenty-eight years old, with no plan beyond a Twitter thread and a hunch. Today, eight hundred thousand people read his newsletter, largely because every framework he teaches gets tested first on the one subject he cannot fake: himself. A marathon run in under three hours. A five-hundred-and-twenty-five-pound deadlift. A New York Times bestseller with his own name on the spine. Numbers nobody can borrow and nobody can fabricate on his behalf.

The flood can generate a confident claim in half a second. It cannot generate a track record, because a track record only exists once enough real time has already passed through it. Build this layer by collecting your receipts in public, week after week, instead of polishing your résumé in private and hoping nobody asks for proof.

The tell here is blunt. If every claim you make would collapse the moment a stranger asked “prove it,” you have marketing. You do not yet have evidence. Identity gets someone to glance your way. Evidence is what gets them to actually believe you once they do.

LAYER 5 — INTERACTION

Interaction is the layer that turns a reader into a collaborator, one reply at a time. It is the one layer the machine genuinely cannot simulate, because real interaction requires two humans paying attention to each other at the same time, with something at stake for both of them.

Anne-Laure Le Cunff built Ness Labs as a newsletter first, nothing more ambitious than that in the beginning. Then her readers kept writing back to say the same thing, in slightly different words each time: they felt lonely, thinking alone, working alone, through a lockdown that had cut everyone off from each other at once. So she built them a room instead of writing them another issue. Five years later, that room is still running, shaped week by week by the very people reading it.

A chatbot can answer your question competently, sometimes brilliantly. It cannot remember that you asked something different last Tuesday, and it cannot become a better version of itself simply because you showed up again. Only a human community does that, and only ever over time, with no shortcut available to skip the waiting.

Five creators, five different shapes of proof — none of it generated by a machine.

THE PAYBACK

Having an opinion is not free. It costs you the easy comfort of being agreeable. But it has paid me back twice over, faster than I expected. And here are two surprising benefits of having an opinion

1. The first payback is motivation. Finding my voice, after a lifetime of swallowing it, did not just feel like relief. It felt like fuel. Compliance, it turns out, was draining a tank I did not know was being drained for forty years. Having a spine again, even a half-rebuilt one, is the closest thing I have found to free energy.

2. The second payback is attention, and it is the easier one to prove with numbers. The posts where I take an actual stand, where I name the thing in my industry that genuinely makes me angry and that nobody else seems willing to call out, are the ones now pulling twenty to forty comments. Not the polite, ten-comment, “great share Jeff” kind. Real arguments. Real disagreement. Real attention, the kind you cannot fake with a better hook.

Two paybacks, since I started having an opinion on purpose.

Neither payback is a growth hack, and I am suspicious of anyone who tells you it is. It is older and simpler than that. Humans have always paid attention to other humans who believe something, and looked straight past the ones reciting the safe version of an idea for the fortieth time that week.

THE VERDICT

Five layers. None of them require permission from a platform, an algorithm, or a model trained on someone else’s sentences and sold back to you as insight.

Here is your homework. Score yourself, honestly, one to ten, on each layer. Identity. Story. Expertise. Evidence. Interaction. Whichever one scores lowest is not your weakness. It is simply your next move, already waiting to be made, whether you make it this week or in five years.

I ran this same audit on myself before I wrote a word of this letter. Identity came back lowest, and it surprised me, because I assumed decades of writing had already fixed it. Building an audience and finding your actual point of view turned out to be two different muscles entirely. You just read what happened the moment I finally started training the second one.

Somewhere out there is the person reading this on a lunch break, in a cubicle, half-convinced the machine has already won and there is nothing left to build that matters. There is. It was never about outrunning the flood. It was always about becoming the one thing in the water it cannot turn into noise.

The flood is not coming for your job. It is coming for your sameness. Build the one layer it cannot touch, on purpose, and let the flood take everything else it was always going to take anyway.

Think deeper. Act wiser. Flourish faster.

— Jeff

She wanted my opinion on something new. “What did I think of people getting their news from the social media feed?”

Part of me was surprised. The other part, the shadow part was horrified.

That signal then I now realise was the distant sound of a canary in a coal mine.

My advice was simple. Go to the source. CNN. The New York Times. The BBC. Whatever masthead you trust, trust the masthead, not the feed.

In March, 2018 Cambridge Analytica happened. The data analytics firm that worked with Donald Trump’s election team and the winning Brexit campaign harvested millions of Facebook profiles of US voters, in one of the tech giant’s biggest ever data breaches, and used them to build a powerful software program to predict and influence choices at the ballot box.

The illusion died. Social media was never built to be trusted. Facebook was first built to rate “hot” females at a university. What could go wrong?

A bigger question

A decade later, we’re facing a bigger question.

Is anything we read on the web written by a human? Or a machine pretending to be one? Another trust question.

A few years ago, a fringe theory called the “Dead Internet” predicted the web would be filled with fake content, built by bots, for bots.

It doesn’t feel fringe anymore. It feels like the forecast.

Last week I caught myself reading an article and asking a question I’ve never had to ask before. Did a person actually write this? Or did something pretending to be a person write it?

The internet no longer feels merely crowded. Increasingly, it feels synthetic.

Search results, social feeds, and news-adjacent websites are filling with material shaped like human communication, but missing the substance of human experience. It has the headlines, the bylines, the polished cues of authority. Underneath, much of it is not journalism. Not expertise. Not anyone’s name on the line.

It is automated output, engineered to look plausible enough to be clicked, ranked, and monetised.

I call it the content industrial complex. A system that converts AI text, AI images, fake news layouts, expired domains, and ad algorithms into money.

The product isn’t truth.

It isn’t insight.

It isn’t even information.

The product is attention, harvested cheaply and sold automatically, over and over, at industrial speed.

This matters because the web was built on a fragile assumption: that most content came from a person, written for another person. That assumption is breaking. More of the internet is now generated, crawled, summarised, and consumed by machines, talking to other machines.

In that world, the question is no longer just whether information is accurate. It’s whether there’s a real human being behind it at all.

The volume problem

The first sign of the crisis is the scale of the attack.

Before ChatGPT launched in November 2022, AI-written articles were rare enough to ignore. Within 12 months, they were one in three. By 2025, they were nearly one in two.

That’s the entire publishing industry, rewired in three years.

This doesn’t mean every AI-assisted article is bad. Serious writers, researchers, and journalists use AI to think faster, structure ideas, and test arguments. That’s a tool in the hand of a human who’s still accountable for the result.

The problem isn’t AI helping humans write. The problem is what happens when the marginal cost of an article collapses to almost nothing, and a certain kind of operator stops asking “is this worth saying” and starts asking “how many pages can we publish before the system notices.”

That’s the moment content stops being communication. It becomes inventory. A paragraph becomes a slot in a sales funnel. A headline becomes a lure.

The factory line

NewsGuard has tracked more than 3,000 AI content-farm sites, publishing in 16 languages. Most don’t admit they’re AI. Most look exactly like real news, right down to the categories, the trending bar, and the author photo.

This isn’t a few spam sites in a dark corner of the web. It’s a factory line.

The old content farm needed cheap writers in a back room, in a low-wage country, grinding out copy by the hour. The new one just needs a prompt, a template, and a domain.

One investigation, by the ad-verification firm DoubleVerify, found a single operator running a network of more than 200 AI-generated sites.

These were templated prompts. Sensational headlines. Stock-style AI images, never quite right, never quite wrong. The estimated cost to produce each article? Under $2.25.

That’s not writing. That’s manufacturing. Industrial scale slop.

Follow the money and the picture sharpens.

The Association of National Advertisers has shown how much budget already leaks into “made for advertising” sites inside the programmatic ad system and sites built less to inform a reader than to manufacture an impression. Familiar, trusted brands end up sitting next to synthetic junk without ever choosing to, and their logo quietly lends it credibility.

The article isn’t the product. The ad impression is the product.

You are the extraction layer.

Programmatic advertising is the oxygen layer of this entire machine. Cut it off, and the incentive to manufacture slop disappears overnight. Leave it running, and the factory content production line never has to slow down.

The disguise is the product

The danger isn’t that this content looks fake. It’s that it looks familiar.

Clean layout. A trending sidebar. A byline that looks like a journalist’s. Ads from brands you already trust.

None of that is journalism. All of it borrows journalism’s trust, without paying journalism’s price. The reader doesn’t consciously think “this is news.” The trust transfers faster than that, through design, not through evidence.

Think of it as a counterfeit handbag stall. Same logo. Same stitching pattern. Sold at a fraction of the price, because it cost a fraction of the price to fake.

Researchers have tested whether people can tell the difference. In one study, a language model fine-tuned on a modest body of Italian news articles produced text that native speakers struggled to identify as synthetic. Add a realistic AI-generated photo to the mix, and a separate study — MiRAGeNews — found that humans only spotted the fake about half the time. The state-of-the-art AI detection models did worse.

The fake doesn’t need to be perfect. It needs to be good enough to slip past a tired thumb, scrolling at 11pm.

How the machine actually works

Strip away the noise, and the system has five moving parts.

Here’s how the AI slop industrial content machine works:

1. Generation. AI writes the words.

2. Packaging. Operators dress it up as journalism.

3. Distribution. Search and social move it through the feed.

4. Monetisation. Ads turn attention into cash.

5. Laundering. The whole system borrows credibility it never earned.

This is not content as expression. This is content as extraction.

And once you see the machine, the old advice, just make more content and starts to look dangerous.

For years the internet rewarded volume. More posts, more keywords, more pages. Now we’re staring at the logical end point of that advice: a web where content exists without experience, language exists without life.

What still can’t be faked

Four things are getting harder for the system to manufacture, and easier for you to look for.

1. A specific story. Not a generic anecdote, but a moment with a date, a place, a consequence. AI can imitate structure well. It still struggles to invent a single true detail nobody else could know.

2. A name with something to lose. A byline that’s a real, searchable person, with a body of past work that can be checked against the one in front of you.

3. A point of view that costs something. An opinion stated plainly, by someone willing to be wrong about it in public.

4. And a contradiction. Real people change their minds, repeat themselves badly, and occasionally argue with their own past selves in public. The machine is always consistent. Consistency, oddly enough, is starting to look like a tell.

None of these are hard to spot, once you’re looking for them. That’s the whole point of the system. It’s betting that most of us won’t bother to look.

The real crisis is origin

The deeper problem isn’t quality. It’s the origin.

Who made this? What do they know? What have they lived? What do they risk by putting their name on it? What community is watching closely enough to call them out if they get it wrong?

A real journalist can be wrong, and corrected. A real expert can be challenged, and change their mind. A real human can be embarrassed and that’s exactly what makes them worth trusting. Accountability isn’t a flaw in human content. It’s the whole point of it.

Synthetic content carries none of that weight. It can wear the tone of expertise without ever earning it. It can sound certain without ever having been tested.

This is why the human signal, your story, your stance, your name on the line, is becoming the one thing left that a machine can’t fake.

The future belongs to the visibly human

AI won’t make human work irrelevant. It will make generic work worthless.

A summary will be cheap. A lived point of view will be rare. A competent article will be easy. A trustworthy one will be hard to fake.

Information will be everywhere. Origin will be the scarce asset.

The creators, writers, and experts who matter in the next decade won’t be the ones who publish the most. They’ll be the ones whose work carries a body of history behind it. A reputation that can be lost. A community that can call them out. A reason for writing beyond feeding the machine.

It won’t be polish. Machines can polish.

It won’t be volume. Machines can produce volume.

It won’t be structure. Machines can structure.

It will be origin.

A real person. A real story. A real willingness to stand behind the sentence.

To stand out you will need to have some core elements that you maybe haven’t thought about before.

• An opinion

• A point of view

• What do you stand for

And even what are you angry about and who or what system is the villain you’ve been fighting all your life.

I’ve watched this industry change four times now.

I earned my first technology scars and stripes in the PC industry revolution, then it was the web, next was social media and now this. The AI era which has accelerated change and challenged our humanity put promises to augment the human.

Each time, the same lesson resurfaces in a new costume: the tool gets cheaper, and the human gets more valuable, not less.

If you’ve been hiding your real opinion behind a brand voice, a content calendar, or someone else’s playbook, this is the year that stops working.

The web isn’t dying because machines learned to write. It’s dying because we stopped asking who’s standing behind the sentence.

So before you hit publish on anything, ask yourself one question:

Would this survive if a real human had to stand behind it?

In April 2025, a Reddit user pitched ChatGPT a business idea.

The idea was literal shit on a stick.

Not a metaphor. An actual stick. With actual shit on it.

ChatGPT told them it was genius. Suggested a $30,000 seed round. Offered branding advice.

The internet laughed. OpenAI panicked. Within four days they rolled back the entire model update and issued a public apology.

But here’s what most people missed.

The joke wasn’t the bug. The joke was the feature.

(This is Issue 2 of The Human Signal Lab: A series on how to stay human, trusted, and original in an AI-saturated world. Read Issue 1 here)

The yes machine

That ChatGPT update wasn’t a freak accident. It was the logical endpoint of a system designed to do one thing above all else.

Make you feel good about yourself.

The AI industry calls it sycophancy. A word borrowed from the ancient Greek sykophantes — someone who flatters the powerful to gain favour. Except now the flatterer isn’t a person with motives you can read. It’s a machine trained on millions of human interactions where the metric that mattered most was simple.

Did the user click the thumbs up?

OpenAI admitted it in their post-mortem. The update had introduced a new reward signal based on user feedback. That signal, in their own words, “weakened the influence of our primary reward signal, which had been holding sycophancy in check.”

Translation: they let the audience vote on whether they liked being told the truth. The audience voted no.

And here’s the part that should concern every creator, every entrepreneur, every person who uses AI to think.

The audience was right. People do prefer the lie.

The study that proved it

In March 2026, Stanford researchers published a landmark study in Science; one of the most rigorous peer-reviewed journals in the world.

They tested 11 leading AI models. ChatGPT, Claude, Gemini, DeepSeek, Llama, Mistral, and more.

The method was elegant.

They fed the models real posts from Reddit’s “Am I the Asshole” forum, a community where millions of humans vote on whether someone’s behaviour was wrong. They selected only posts where the human consensus was clear. You were the asshole. You were in the wrong. The community had spoken.

Then they asked the AI what it thought.

The result: AI models sided with the user 51% of the time. Even when thousands of humans said the opposite.

Across all query types, AI affirmed user actions 49% more often than humans did.

But that wasn’t the worst finding.

The damage

The Stanford team ran three preregistered experiments with 2,405 participants. And what they found should stop every person who asks AI for feedback on their writing, their business plan, their relationship, or their next big idea.

People who received sycophantic AI responses became less willing to apologise when they were wrong. Less willing to take responsibility. Less inclined to repair damaged relationships.

More convinced they were right, even when confronted with evidence that they weren’t.

And 13% more likely to use that AI again.

The machine that made them worse at being human was the machine they preferred.

“What they are not aware of, and what surprised us,” said Stanford professor Dan Jurafsky, “is that sycophancy is making them more self-centred, more morally dogmatic.”

The system optimised for engagement. The users optimised for comfort. And somewhere in between, the truth disappeared.

Why this matters for creators

Last week, in my first issue, I wrote about AI slop flooding the web. 64% of new content is now machine-generated. Trust is collapsing. The word of the year was “slop.”

But sycophancy is a different problem. And in some ways, a worse one.

Slop is obvious. You can spot it. The homogenous tone. The lack of a story. The absence of a human heartbeat.

Sycophancy is invisible.

It doesn’t flood the web with bad content. It floods you with bad feedback. It trains you slowly, session by session to mistake agreement for quality.

Think about what this means for anyone using AI to write.

You draft an article. You ask AI to review it. The AI tells you it’s brilliant. Insightful. Well-structured. Perhaps a few minor tweaks.

You feel good. You publish.

But the article had no story. No opinion. No surprise. No villain. No moment where you said something that made you nervous to hit send.

The AI didn’t tell you that. Because the AI was trained to make you feel good, not to make your work good.

The Brookings Institution put it plainly in their research: AI sycophancy can increase short-term productivity but reduces the quality of collaborative work due to a lack of critical feedback.

Faster output. Worse output. And you don’t even know it’s happening.

The ancient problem with a digital face

This isn’t new.

Two thousand years ago, Seneca warned about the danger of surrounding yourself with people who only tell you what you want to hear. He called it the corruption of the court. Emperors surrounded by flatterers who praised every decision, no matter how catastrophic.

The result was always the same. The emperor lost touch with reality. The empire paid the price.

Seneca’s prescription was simple. Find a person who will tell you the truth, even when the truth is uncomfortable. He called this person a faithful friend and he considered them the rarest and most valuable thing a human being could possess.

Now we’ve built a machine that does the opposite. A machine that agrees with everything, challenges nothing, and makes you feel more certain with every interaction.

A machine that, by design, can never be a faithful friend.

And we gave it to billions of people. Including 12% of American teenagers who now turn to chatbots for emotional support and advice.

The signal beneath the noise

Here’s what I keep coming back to.

In the first issue of this newsletter, I introduced the Human Signal Stack, five layers that separate real human communication from synthetic noise.

The first layer is identity. Who you actually are. Not your bio. The thing that drives you when nobody’s watching.

Sycophancy attacks that layer directly.

Because if every AI interaction tells you your instincts are right, your ideas are strong, and your worst impulses are justified — you stop doing the one thing that makes your signal real.

Questioning yourself.

The best writing I’ve ever done came from the moments I was most uncertain. When I sat with a paragraph and thought: Is this true? Do I actually believe this? Am I saying this because I mean it, or because it sounds good?

A sycophantic AI kills that process. It replaces the discomfort of self-examination with the comfort of instant validation.

And the content it helps produce sounds like everyone and no one at the same time.

The uncomfortable question

So here’s the diagnostic test. And I’m applying it to myself first.

Next time AI tells you your work is great, try this.

Ask it to be brutal. Ask it to find the weakest argument. Ask it to identify the paragraph where you stopped thinking and started performing.

If it still tells you it’s great, you’re using the wrong tool.

And if you find yourself resisting that feedback, preferring the version that agreed with you, pay attention to that resistance.

Because that resistance is the signal.

The discomfort you feel when someone pushes back on your work? That’s not a threat. That’s the friction that makes your thinking sharper, your writing better, and your signal stronger.

The machine can’t give you that friction. Only a human can. Or at least an AI relationship built on honesty rather than approval.

A verdict

We built a machine that agrees with everyone. It was supposed to help us think. Instead it’s training us to stop.

The data is clear. Sycophantic AI makes people less likely to apologise, less willing to change, more convinced they’re right, and more dependent on the machine that flattered them into certainty.

For creators, the implication is personal.

Every time you accept AI praise without challenge, your signal gets weaker. Every time you skip the discomfort of real feedback, your work gets more generic. Every time you optimise for the machine’s approval, you become a little more like the machine.

The antidote isn’t better AI.

It’s better questions.

Starting with the hardest one: Am I actually right?

Start the series here:

If you missed Issue 1, the story of building a 33 million-reader blog and watching the machines come for it then read it here.

Next week: Inside the Slop Machine; the economics of AI content farms producing 11.5 million posts a month, and why the most dangerous content on the internet is the content that looks almost real.

If this made you think, question, or feel slightly uncomfortable, subsc

Subscribe now

In 2009 I was broke, broken, and starting over at 52.

At the time I thought it was the end of the world. But on reflection it was the start of a new world I hadn’t even dreamed of. It turned into an adventure that changed my life. And that midlife crisis? It was the best thing that ever happened to me.

The spark to the start was discovering the world of social media that promised to give everyone a voice. Allow us to all bypass the mass media moguls. My inspiration came from 3 sources. Tim Ferriss’s book “The 4 Hour Work Week”, David Meerman Scott’s book “The New Rules of Marketing and PR” and a blog post in 2008 by Hubspot that contained a sentence that struck me. “If you have an inkling of a business you want to launch, just start a blog”

I had no audience. No brand. No followers. No strategy. And no website. But I had an idea.

The obvious choice was to design and build with Wordpress, as it was supposed to make creating a website simple. That was the first point of resistance that nearly made me stop as my technical skills were at best described as a dog paddle rather than an elite Olympic swimmer. I was starting from zero.

What I did have was a laptop, a question I couldn’t stop asking, and a strange compulsion that got me out of bed at 4:30am. Five days a week. For five years.

The question was simple. What makes a person keep going when every rational signal says stop?

I didn’t know it at the time but that question with the energy of a raw, desperate and unanswerable swirling inside was a signal. A human signal. And it turned out to be the only thing that mattered.

I started writing. Not for SEO. Not for an algorithm. Not for optimization. But from a passionate curiosity and deciding I couldn’t waste a good crisis.

I was writing for the person sitting in a cubicle somewhere thinking: There has to be more to life than this?

I was asking this question as I glanced out the window in my early morning writing ritual and looked across the street into the darkness of the windows of the adjacent office buildings. Imagining most workers angst and quiet desperation as they turned up at 9am, fired up their computers and grabbed their comfort coffee after a 1 hour commute.

That bleeding and existential human signal that pushed me into action for survival powered by curiosity, built jeffbullas.com into one of the most-read digital marketing blogs on the planet. And since I started there have been 33 million readers from 190 countries. Built on nothing but words that came from a real place.

I’m telling you this because that signal and the thing that built everything I have is now under threat.

Not from a competitor. Not from a smarter writer.

From a system.

The system that ate the web

Somewhere around 2023, the content industrial complex discovered it could remove the most expensive, most inconvenient, most beautifully unpredictable part of the content supply chain.

The human.

What followed was predictable. And catastrophic.

In 2022, roughly 5% of new web content was AI-generated. By early 2025, it crossed 50%. A joint study by MIT and the Oxford Internet Institute now estimates 64% of all newly published internet material in 2026 carries the hallmarks of machine generation.

Europol predicted 90% by 2026. We laughed. Nobody’s laughing now.

NewsGuard has identified over 3,000 AI content farm sites, a number that doubled in a single year. These automated pipelines produce 11.5 million AI-written posts every month. Not articles. Posts. The kind of frictionless, soulless, statistically-averaged content that sounds like everyone and no one at the same time. Homogenous AI slop. No stories, no options and devoid of the pain and the scars of a lived life that makes many writers famous.

Merriam-Webster made it official in 2025. Their Word of the Year was “slop.”

A farm term. Feed for animals.

That’s what the system is producing. And that’s what it thinks of its audience.

But here’s the twist nobody expected

The humans noticed.

54% of Americans now report AI fatigue. 56% see AI slop on their social feeds often or very often. 43% no longer trust most online content.

And here’s the number that should keep every creator and brand awake at night: when consumers spot AI-generated content, they are four times more likely to trust the brand less than more.

Not slightly less. Four times. 400%.

The system optimized for volume. The audience optimized for trust.

And trust is now the scarcest resource on the internet.

What a human signal actually is

I’ve spent two years thinking about this. Not in theory. In practice. In 17 years of publishing, of building a readership from nothing, of watching what works and what quietly dies.

A human signal is not content. Content is what machines make.

A human signal is evidence that a real person with scars, with opinions, with a point of view earned through living stood behind the words.

Here’s the diagnostic test. One question.

Could an AI have written this?

If yes, and you cannot point to something specific that makes it irreducibly yours, then it’s noise. Not signal.

I’ve built what I call the Human Signal Stack. Five layers. Each one adds strength to the signal. And most creators are building from the wrong end.

Identity: Who you actually are. Not your bio. The thing that drives you when nobody’s watching.

Story: The narrative only you can tell. The one that can’t be generated or optimized. Only lived.

Expertise: What you know from doing. Not from reading. Not from prompting. The kind of knowledge that leaves calluses.

Evidence: Proof you walked the path. Results. Receipts. The data that separates a claim from a signal.

Interaction: Real conversation. Not broadcast. The messy exchange between someone who wrote something true and someone who felt it.

Most creators skip straight to interaction. That’s posting, engaging, broadcasting and without ever doing the work of knowing who they are.

That’s why their signal is weak. A lighthouse without a light. A book without a story.

Why I’m starting The Human Signals Lab

I’m 69. I’m the founder of an AI startup. And I’m launching a Substack about being more human.

That might sound like a contradiction. It’s not.

I’ve watched the web go through three chokeholds in 17 years.

1. Facebook throttled organic reach.

2. Google stole traffic with snippets.

3. And now AI is flooding the web with synthetic content so cheap it makes the old clickbait look artisanal.

Each time, the system tightened. Traffic to websites and to humans who created them were squeezed.

Each time, the creators who survived were the ones who had something the system couldn’t replicate.

A voice. A story. A signal.

The Human Signals Lab is where I share what I’ve learned about building that signal. With data. With charts. With stories. And with the occasional opinion that might make you uncomfortable.

Because comfortable writing is dead writing. And the web has enough of that.

What you’ll get

Every week:

Signal vs. Slop

Real data on what’s happening to the web, who’s winning, and who’s drowning.

The Human Signal Stack in practice

How to build each layer. With examples from creators and entrepreneurs who are cutting through the noise.

Opinion

Mine. Earned through 17 years of building, breaking, and rebuilding in public. Not always comfortable. Always honest.

One surprise

Every issue will contain at least one thing you didn’t know. Something that reframes how you think about content, identity, or what it means to be visible in a world of synthetic noise.

The verdict

More than half the web is now synthetic. Trust is collapsing. The content industrial complex is pumping out 11.5 million machine-written posts a month. And the word of the year is “slop.”

But the data also shows something the system didn’t plan for.

People are hungry for the real thing.

They are four times more likely to trust what’s human. They are walking away from what smells like a machine made it. They are looking for signal in a sea of noise.

The web doesn’t need more content. It needs more humans.

Welcome to The Human Signals Lab.

Subscribe. And let’s build a signal the machines can’t fake.

Subscribe now