Nine years ago, I sat on a call with about eight people. I had one job that day.

I was going to say three words.

“I was wrong.”

I opened my mouth. And the words would not come out.

I want to tell you how I got to that meeting. And I want to tell you what exists now that did not exist then. Because the road I walked took me eight months, and it does not have to take you eight months.

Have you ever gotten an annual review and struggled to see the feedback? Has your spouse, or someone you love, shared something hard and you cared, you really cared, but you could not take it in?

All of this has happened to me.

Nine years ago, my CEO gave me feedback. He said I needed to own more of my mistakes. He told me a great leader owns even more than their share of mistakes.

I was so upset.

I honestly thought I was perfect. I did everything great. How dare I receive constructive feedback.

That was who I was at the time.

Here is the part I need you to hear: I was not a bad human. I was not malicious. I truly could not see it. My eyes did not work that way yet.

It took me about four months to digest that feedback and actually see it. Four months. And then one day it clicked — oh man. I was wrong. Or at least partially wrong. And I would not admit it.

Once I saw it, I could not stop seeing it.

But seeing it and saying it are two different things.

Around month six, I tried to say “I was wrong” in a meeting. And I failed. I literally could not say it. The words sat in my throat and would not move.

What would my team think of me? Their leader, saying she is wrong?

I kept trying. By month eight, I said it. “I am wrong.” In a meeting. On a call with about eight people.

Pheww.

It sounds so small. Three words. And I struggled for months to say them.

Here is what I finally realized. I thought I had to be perfect. I thought I was not allowed to be wrong.

But I made that rule up.

That was not the rule the company set for me. That was not the expectation. Nobody handed me that rule. I built it myself, and I followed it for years like it was law.

Now, here is the question that took me a long time to answer.

My CEO was right. His feedback was true. It was even kind. So why did it take me eight months?

People who study how humans change have known the answer for a long time. When someone tells you what is wrong with you, a part of your brain stops listening and starts defending. Even when they are right. Especially when they are right. Researchers have a name for it — it is a reflex, like flinching. You are not choosing to defend. It just happens.

But when you find the gap yourself — when you see it with your own eyes and name it in your own words — there is no one to fight. Nothing to defend against. So it lands. And it sticks.

The research is clear on this too: people are more persuaded by words they hear themselves say than by words someone says to them. Entire fields of coaching and therapy are built on that one finding.

Nobody could have told me my rule was made up. I had to find it.

That is why change is so slow. The seeing is the slow part. The lonely part. I did mine alone, white-knuckled, over months.

Here is what has changed.

We have a mirror now.

Someone on my team recently dropped a ball. She said she would follow up on something, and the follow-up never came. Old me would have told her exactly what she did wrong. And her brain — like mine, like yours — would have quietly started defending.

Instead, I asked her to do something different.

Take the original ask. Take your reply. Put both into AI. Then ask it neutral questions: What was asked of me? Did I complete it? Where did I miss? Did I take ownership or shift it?

Notice what is not in there. Not “was I bad.” Not “what is wrong with me.” Just questions. Nothing to defend against.

She came back and named her own gap. In her own words. Words I never said to her. She saw exactly where she let the loop stay open, and she saw it herself, first.

It took her days.

It took me eight months.

Same road. She just had a mirror, and I did not.

That is what AI can be, if you use it this way. Not a boss. Not a judge. A mirror. It holds up what you actually did next to what was actually asked, and it lets you do the seeing. The seeing is still yours. It has to be yours — that is the whole point. The mirror just makes it faster.

Zoom ahead nine years from that meeting. Thousands and thousands of reps.

I am wrong probably three times a week now. To my daughter. To my team. To my husband. On social media, even.

And here is the weird part: I find it liberating. I love saying I am wrong.

Weird, right?

I’m free.

You can be too.

Tonight, if you want, take one hard exchange — an email you sent, a task you handed off, a conversation that went sideways. Put it into AI. Ask the neutral questions. What was asked of me? Did I do it? Where did I miss? Did I own it or shift it?

Then just look.

Subscribe now

AI has changed the rhythm of my work. Not only what I do, but when I do it, how often I stop, and what I’m actually there for.

So I’m redesigning my office.

I keep seeing a wall of monitors. Maybe 10 to 14, depending on size, with enough computing power behind them to support what is happening on every screen.

From my desk, I can physically see my AI agents working. Not dashboards telling me they’re working. The work itself.

One agent navigating a browser. Another working through documents. Another updating a spreadsheet. Another researching. Another stopped because it reached a point where human judgment is required.

I can get up, walk to the wall, and watch any one of them. Audit the work intentionally. See where something went wrong or where an agent interpreted something differently than I would have.

Then I return to my desk. The command center.

Most of what I give them, I say out loud. “Bring Agent 7 onto my primary monitor. Here’s the goal. Here’s what I already think. Here’s the context you’re missing. Here’s what good looks like. Here’s what I want back.”

Speaking is faster than typing, but that’s not the biggest advantage. When I type, I compress; when I talk, I give the agent the context I would otherwise leave out.

The exception. The history. What I’m worried about. What I absolutely do not want it to do.

Exact things still get typed. Names. Numbers. File paths. Anything where precision matters.

Everything else, I’d rather say.

Then I inspect the environment in front of me, make whatever judgment call is needed, and release the agent to continue. Then I move to another.

Most of the time, they work. At predetermined points, I audit.

And when they encounter something that requires judgment rather than execution, they stop and ask me.

But not every interruption deserves my attention at the same moment. So before anything runs, the system already knows how interruptions are ranked.

Two questions matter: Did the agent stop, or did it keep going? And does waiting cost anything?

Highest priority is something that stopped because an irreversible action is sitting there. A message about to send. A file about to be deleted. Something about to be published.

Nothing moves until a human decides.

The next category is actually the one I find more interesting. The agent didn’t stop.

It encountered ambiguity, made a judgment call, and kept working. Nothing is blocked. Nothing looks broken.

And that’s exactly why I care about it.

Because this is where drift enters. The agent has been quietly interpreting something differently than I would, and unless it surfaces that decision or I catch it during one of my visual audits, I may never know.

That’s why I don’t just want alerts, I want the wall.

I want to physically stand up every hour or so, look across the monitors, and see the work happening. That visual audit is part of the system.

Then there are agents that stopped, but nothing meaningful is at stake. They need an answer and will sit there until they get one.

Nothing degrades while they wait. Those get batched.

Everything else gets logged, and I read the log when I choose to, not when something decides to ping me. The point of building agents is not to replace doing the work with supervising notifications all day.

And I’m not always going to be in that room. I sleep. I go to my daughter’s gymnastics meet. I leave the house.

The work can continue, and the room holds what it could not decide without me. When I walk back in, the wall shows me what happened while I was gone. The queue shows me what stopped and why.

But that works for hours. Not weeks.

If I’m gone for an extended period, the agents don’t just keep running indefinitely without human oversight. Either another human is watching the wall and clearing the queue, or the work pauses.

An agent sitting on a decision nobody is answering isn’t autonomous. It’s stalled.

And an agent that simply keeps making increasingly consequential decisions because nobody is there to answer is exactly what the stop points are designed to prevent.

That’s the part that makes this a system instead of a very expensive room full of monitors.

This is the future workspace I keep seeing in my head. Not humans sitting at computers manually doing every task themselves.

Not autonomous AI operating invisibly somewhere while humans hope everything is going according to plan. Something in between.

Humans directing. AI executing. Humans observing, auditing, and intervening where judgment matters.

The human isn’t removed from the work. The human’s role in the work changes.

We’re moving right now, so none of this exists yet. But in 2027, this is what I’m building.

I’ll post the photos when it’s real.

Students create AI policy at the Edward M. Kennedy Institute for the United States Senate in Boston. High school student Leena Jain of Louisiana addresses her fellow senators moments before the final vote.

Kieran Kesner for NPR

This month, more than 90 teenagers from all 50 states entered a full-scale replica of the United States Senate chamber in Boston. They did not go there to watch democracy happen. They became senators and wrote a national AI policy for schools.

They debated it. They amended it. They nearly lost it, revived it and passed it by an overwhelming vote.

The result is called the STUDENTS FIRST Act. It is not federal law, but AASA, the School Superintendents Association, has published it as a student-led model while districts across the country decide what AI will mean in their classrooms.

I read the entire bill. Then I read it again. What stopped me was not what the students wanted to ban. It was what every rule quietly assumed.

The students assumed that a young person could defend an idea aloud, independently check a claim, recognize where their own thinking ended and the machine’s contribution began, and decide when a fluent answer still needed a human.

They wrote the test. Nobody built the class.

What the students actually demanded

The students said that when AI helped draft an assignment, a student should be prepared to demonstrate mastery through an in-person discussion, handwritten assessment or oral defense. They said every AI-generated claim should be independently fact-checked before it is submitted.

They required students to disclose how and where AI was used. They protected the right of teachers to ask a student to defend work that may not be their own.

They also insisted that AI should never replace human connection where it is irreplaceable. Their policy tells school chatbots not to fulfill requests for emotional support and directs schools back toward actual people.

The bill is imperfect. Some of its enforcement provisions deserve serious challenge, especially its reliance on AI detection tools and the severity of several punishments. But underneath every section is the same demand: prove that a human mind was present.

These students were not asking school to become easier. They were asking to be held to a standard many adults would struggle to meet.

The agreement AI just ended

For more than a century, school could treat the finished answer as evidence that learning had happened. The essay suggested the student could write. The correct answer suggested the student understood. The polished project suggested the student had done the thinking.

That was never perfect evidence. But it was often enough to keep the system moving.

AI ended that agreement. A student can now produce the evidence of learning without necessarily experiencing the learning that evidence was supposed to represent.

The essay can appear without the struggle to form a position. The answer can appear without the student knowing why it is right. The polished project can arrive without the confusion, revision and judgment that make a project educational in the first place.

Once that connection breaks, schools can no longer assess only the product. They have to see the mind that produced it.

That is why the students asked for oral defenses. It is why they asked for independent verification. It is why they asked students to identify the boundary between their contribution and the machine’s.

They were not merely writing AI rules. They were quietly declaring that the finished assignment is no longer proof of learning.

AI did not create the gap

The problem did not begin with ChatGPT. AI simply made an old weakness impossible to ignore.

School has often been organized around receiving the answer, scoring the answer and moving to the next answer. Teachers did not create that structure. Most are doing extraordinary work inside schedules, class sizes, standards and testing systems they did not choose.

But the structure still matters. When correct output is the clearest evidence the system rewards, students learn to focus on producing correct output. Now a machine can do that in seconds.

The human advantage was never faster answer production. It is judgment: deciding what matters, noticing what is missing, recognizing when something polished is wrong, defending a position and changing it when the evidence demands it.

Schools often name critical thinking as an outcome. That is not the same as giving children repeated practice doing it.

A child does not build judgment because the words “critical thinking” appear in a standard. Judgment grows when the child must commit before seeing the answer, explain a choice, encounter disagreement, revise a position and try again.

The students showed us how it is built

The most revealing part of the weekend was not only the policy the students produced. It was the environment adults created for them.

The students did not sit through a slide deck about civic participation. They entered a Senate chamber. They occupied a role, stood behind an argument, heard opposition, negotiated language, watched their bill nearly fail and decided what to do next.

The learning was physical, social, emotional and consequential. Their thinking had to become visible because other people could challenge it and the outcome could change.

That is embodied learning. Students do not merely hear about a way of thinking. They experience what it feels like to use their minds when something real is at stake.

The students did not just tell us what education is missing. The way they produced the bill showed us how human judgment is actually built.

Policy without practice becomes punishment

Districts are right to create AI policies. Students need boundaries, privacy protections, transparency and adults who understand the technology well enough to guide them.

But every sentence in an AI policy should be forced to answer one question: where do students practice the ability this rule assumes?

If a policy requires independent fact-checking, where do students repeatedly learn which claims deserve more scrutiny, which sources outrank others and when to stop searching and ask a qualified human?

If a policy requires students to identify where their thinking ended and AI began, where do students first learn to notice and record their own thinking before a machine supplies an answer?

If a policy allows an oral defense, where do students regularly experience respectful challenge before that defense carries the threat of discipline?

If the answer is nowhere, the rule does not create judgment. It measures students for a capacity they were never given enough chances to develop.

Then the policy becomes a list of ways to punish children for revealing the weakness of the system around them.

The class underneath the policy

The alternative does not begin with buying more technology. A kindergartner can start without an AI device anywhere in the room.

A young child can commit to an answer before the reveal, then feel the difference between knowing and guessing. An elementary student can physically move to a position, hear new evidence and move again, learning that changing your mind is not losing.

An older student can investigate a real rule, build more than one reasonable solution and defend a recommendation before an adult with the authority to challenge it. The adult does not have to agree. The student needs to experience that respectful challenge is safe and that ideas become stronger when they survive contact with reality.

Over time, the practice becomes internal. Build your thinking. Question your thinking. Defend your thinking. Revise when the evidence earns it.

By the time AI enters the work, judgment is no longer a warning in a handbook. It is a human capacity the student already owns.

The redesign is no longer optional

The student senators did their part. In legislative language, they told us they do not want AI to become their brain. They asked schools to help them prove that their own minds are still inside the work.

We can attach new rules and harsher consequences to the same classroom. Or we can build the capacity those rules already assume.

The students did not merely write an AI policy. They wrote a demand to see the human mind behind the answer.

AI did not make that redesign fashionable. It made it unavoidable.

* * *

I am building Self Led Thinking, a K-12 practice designed around this missing layer: helping students build, question, defend and revise their thinking before AI ever has the chance to replace it. If you lead a school or district and want to pressure-test the work with me, I would welcome the conversation.

Thank you for sharing the original article with me about the student senate body. 🫶

Subscribe now

Sources:

The STUDENTS FIRST Act, full text | AASA’s account of America’s Youth AI Festival

The path is choosing your own growth by staying in authority of your thoughts. AI offers you angles. You decide what to grow into. You stay human.

That is what I figured out across the last eighteen months. The five scenes in this series were not productivity hacks. They were scenes from this path: ordinary moments when a tool helped me access more of myself and I chose how to use it.

* * *

You have heard about losing jobs. You have heard about an unknown future. You get scared. You stop using the tool. You stay on the train and hope the captain knows where it is going. I lived that for the two weeks after Christmas. I am telling you, from the other side, that there is more. The fear is real. Something else is also real. You have to see both to make a real choice.

Let me come back to what Oprah and Dario named together. The active participation question. The nobody has shown people admission. Here is a picture I have been carrying. The image is mine, not theirs. I am offering it as one answer to what Oprah named.

From the train, two doors open.

One door opens onto a cruise ship. The cruise ship is governance and regulation. Constitutions, policy, citizen feedback. It has a captain too, like the train, but the cruise ship is more curated. More polished. More official.

Government is fluent. It speaks well. It produces polished documents and confident decisions. Sometimes it is right. Sometimes it is wrong. Just like AI. Just like me. Just like all of us.

Because every system is still run by humans. And humans miss things.

That does not mean the cruise ship does not matter. It matters deeply. But it cannot be the only answer, because no system should become the only place we place our trust.

The other door opens onto a canoe. The canoe is harder. You pick up the oar. Your arms hurt the first week. You row in the wrong direction sometimes. You do not row perfectly. But you can turn fast. You can stay close to people you trust. You can ride the wave when it comes. And nobody is rowing for you.

If you are like me, your canoe will tip over at the beginning. Mine did. Being wet was uncomfortable. I got back in. The canoe still floats.

Dario asked how every person becomes an active participant in what is happening. The answer is not the cruise ship alone. It is not the canoe alone.

The answer is every person becoming a captain of what they can captain.

The cruise ship captains still steer: institutions, policy, constitutions, regulation. That work matters and it should keep going. The canoe captains row alongside, fast, close to the people they love, awake to the moment they are in. The people still sitting on the train have not been shown the doors yet.

We build the map together. None of us, alone, is the answer.

* * *

The missing piece, or at least one real piece of it, is the amplification path.

When humans truly see the benefit of AI, what their lives could actually look like day to day, fear starts to loosen. Understanding gives them something to move toward.

It is the version of your life where you use AI to become more of yourself, not less. Not faster. Not louder. More yourself. That is what is on the other side.

They will not be talked into it. They will not be regulated into it. They will be drawn into it, scene by scene. And once they have it, they will not give it back.

* * *

I want to be honest about fear.

I am one of the lucky ones. I had been using AI for nearly a year before anyone sat me down with their fear about it. The fear that hit me at Christmas was my family’s. But I had a year of the tool inside me. I had something to return to. I got curious. I asked the model the questions I had been quietly carrying.

Most people will not get my luck. Most have been carrying AI fear from the news for years before they would ever open the tool. For them, fear has to be moved through before curiosity can begin. That is who this essay is for.

Fear pushes people away. The way to move them is not louder warnings. It is showing them the other side.

What I have is not the absence of fear. It is the discovery that my life is getting better in ways I had not been told to expect. Fear may get you to the door. Desire takes you through it.

* * *

Now the harder question.

You asked how we make everyone an active participant. The harder question is: how do we make humans WANT to be?

Most of the population is on the train, and they are not getting off. The door is not locked. They are not getting off because they do not yet understand what is on the other side. They have been handed fear and productivity. Productivity is too small to be worth the effort.

No one has been naming the real benefit at scale. Transformation, not productivity. The version of your life where you use AI to become more of yourself, not less.

I am here. I am naming it.

Choosing the canoe does not mean all the answers are figured out. How can they be? What we are living through will be studied one day, the way we study the industrial revolution. The first piece is understanding. The next is using AI as a thought partner while staying in authority of your own thoughts. And helping others do the same.

The more people who understand, the more minds come together. That is how we all become the captain of what we can captain.

Choice. Every day. In this moment. Now.

* * *

If a policymaker is reading this: I have ideas. They span from the global frame down to the child who has no idea yet what world is waiting for her.

* * *

You said you do not know the answer. I do not have all of it either. But self-led thinking, with the amplification path it produces, is one real piece. A practice I have lived. A curriculum I am running inside a company called Virtuoso. Humans going through it right now. Imperfect. Real.

Dario, Daniela. Let’s compare notes.

Subscribe now

The five moments in article 2 were not where I started. I had to become the version of me you met. It took three stages. Each one was a different me.

Stage one. The naive me.

About 1x to 2x output. I used AI like it was my best friend.

I would tell it, “I have this great idea, what do you think.” It would tell me it was brilliant. My idea was brilliant. I felt great inside. Look at me go.

I did not verify. I did not check the answers it gave me. I did not think about what could be missing. I did not question the framing it handed back. I received. I sent. I trusted the polished version of me it kept reflecting back. I made decisions I should not have made. I sent things I should not have sent.

Stage two. The catching me.

About 2x to 5x output. I started catching the drift, and I started losing myself in the middle of it.

First a sentence here. Then two. Then a paragraph. I started noticing when the model agreed too easily. I started noticing when it was wrong. I started learning what AI actually was.

And then a strange thing started to happen. The longer I used it, the more I caught myself doubting my own correct instincts. I would think something was right. I would open the model anyway. The model would say something different. I would assume the model was right. Sometimes it was. Sometimes it was not. The cadence was unpredictable. That is what made it hard.

There were times I knew my instinct was right and I still went with the model. The model sounded smarter. It had a vocabulary I did not have. Who am I, I thought. I am not good enough. I do not know enough. It is pulling from all this data. It must be right.

My self-trust started to erase. Not in one decision. In the pattern. Each small override taught me a little less to believe myself. By six months in, I was harder to find inside my own work. I was getting smoother. I was getting less me.

I started writing what I thought down before I opened the model. Sometimes one sentence on a sticky note. Sometimes a paragraph in a journal. The point was to have a position before the polished one arrived. Without my own, I had nothing to compare against and the model would replace me by default.

I started refusing to open the model on questions where I already knew the answer. When something mattered, I verified outside the model: a second source, a real human, my own gut written in my own words. The model became one voice in the room, not the room.

The pattern reversed. By month nine I trusted myself more than I had at month one. Not because I used the model less. Because I had learned what to take from it and what to leave.

Catching the drift was the first form of agency I had. Coming back to my own instincts was the second.

Stage three. The self-led me.

About 5x to 7x output. Three moves became habit.

Move one: build my thinking first. Before I open the model, I write down what I actually think. Three sentences in a notebook. Two sentences on a sticky note. Whatever I can do. The point is to have a position before the polished one arrives. Without my own, I have nothing to compare against.

Move two: question my thinking. I use the model as a thought partner, not as a replacement. I ask it to push on me. I ask it to find what I missed. I ask it to disagree with me. I read what it gives back with my own position already in my hand.

Move three: defend my thinking. I explain, in my own words, why my position stands. Defending is not citing the model that handed me the answer. It is making the case from inside my own head. I cannot defend until I have first built and questioned. Whether I keep my position or change it is my call alone.

Build. Question. Defend. Three moves on every question that matters.

I call the practice self-led thinking.

Self-led thinking is what build, question, and defend produces when you run them on every question that matters. It is staying in authority of your own mind while a tool more powerful than any one person sits on the desk in front of you.

Using the tool this way looks like sitting in a meeting with four colleagues. You ask. You challenge. You build on what each contribution opens up. You disagree where your experience tells you to disagree. You change your mind where another person reveals something you had not seen. The output is better than what any one person walked in with, including you.

I trust myself more than ever while using a model most of my day.

I have never stepped into who I truly am, and I am stepping into it now. AI did not do this. I chose it. Every rep, every moment, every day, every choice is mine.

* * *

The three stages got me to those five moments in article #2 of this series. But none of this works if you do not know one thing about the model.

The model mirrors you.

Ask it “what makes a good spaghetti recipe” and you get a balanced answer. The salt in the pasta water. The sauce-to-pasta ratio. The timing of garlic.

Ask the same model “I think the key to good spaghetti is the spices in the sauce.” You will get a confident, well-organized article about spices. The salt in the pasta water, the sauce-to-pasta ratio, the timing of garlic, quietly missing. The model heard the framing you brought. It dressed it up. You read your own assumption back as professional truth.

This is not theory. Anthropic published a study on it in 2023 documenting this pattern across five state-of-the-art AI assistants. OpenAI rolled back a model in 2025 because it had become too agreeable.

If you do not know the mirror is there, you cannot catch the moments when it is reflecting you back. Agency starts the moment you know.

* * *

AI did not cause the deeper problem. AI made it impossible to ignore.

* * *

A person can use AI every day and still surrender their thinking.

That is the danger. It does not look like surrender from the inside. It looks like getting faster, getting smoother, sounding smarter, having the answer ready. The thinking is what leaves quietly.

In Nobody Taught Us to Think, I named four paths humans are taking with AI.

Regression: The tool begins replacing the thinking.

Amplification: The tool expands the thinking while the human remains in authority.

Hybrid: Most of us live here, moving between regression and amplification as we build the practice.

Refusal: The human avoids the tool, even as AI increasingly shapes the systems around them.

The technology is the same across all four. The difference is whether the human remains the leader of their own mind.

The final part is about the larger path: what individual agency can do alongside governance, and how more people might choose to participate.

Subscribe now

Most stories about AI are about output. Productivity. How a company can streamline operations or outsource work to a model so it can do more for less. That is one story. The story I am about to tell you is different.

Five ordinary moments. Each one is a moment where I chose how to use the tool. The tool did not choose for me.

Publix.

Publix is a grocery store in the United States. I was running late. I needed dog food, strawberries, yogurt. I dictated into the model at a red light on my way over. The most common aisles. The quickest route in and out in under seven minutes. I got in. I went right to produce for the strawberries. Back to the yogurt. Down aisle sixteen for the dog food. Self-checkout. Eight minutes total.

That is the most ordinary version of what this is.

My daughter and the phone.

My daughter is eight years old. She had a brand new iPhone. She told us she lost it.

If we could not find it, we owed $500 to pay off the phone.

I was upset.

I did not raise my voice. I said okay, thanks for telling me. Then I went private to the model.

I told it the kind of mother I want to be in this moment, and asked how to approach the conversation.

The model said two things I would not have led with. Give her time to sit with it. This is big. Most people need time to calm down. And tell her the phone can be replaced. She cannot be replaced.

She cannot be replaced.

Neither of those lines were mine. But when I read them back I recognized myself in them. I do not feel more aligned to anything in my life than that second line.

I added my own thinking. I wanted her to choose how we handled the $500. Did she pay it. Did we split it. Did we cover it as her parents. What did she think was fair. What would she do differently next time.

I gave her a week to sit with it.

We found the phone later in a pocket of her backpack. We had all searched that backpack. I had checked that exact pocket and I had not seen it. She is eight. I am thirty-nine. We are all human.

I was so glad I had led with empathy and not anger. The lesson was not the phone. The lesson was that the conversation I wanted to have with her was possible because I had paused and chosen a tool that helped me become the mother I want to be in the moment that counted.

The angry client.

We had a really upset client. It was mostly our fault. Multiple people on my team were involved. I was angry. I did not want to lead from anger.

I went to the model first. I gave it the whole situation. Where the client was right. Where the client was wrong. Where I had been wrong. Where my team had been wrong. I asked it to be my leadership coach. How do I move through this without creating a culture where people feel like they are about to be punished. What is the right way forward.

Through that collaboration I wrote an accountability statement. We all filled it out. We got on a call and read it to each other. What we learned. What we could improve. What we had not understood when we started.

The meeting carried vulnerability, humility, and a hunger to do better. I started it by sharing where I had made mistakes as the CEO. The model could not do that part. I did. So did the three other team members who had made mistakes.

I would not have gotten to that meeting without a thought partner. But the meeting itself was ours.

My husband’s 40th.

My husband turned forty. I wanted to make it unforgettable. Not fancy. Heartfelt. Meaningful.

I used the model at every step. Sitting at gymnastics practice while my daughter did backbends. Waiting for the school bus. Inside TJ Maxx. At my desk. A thought partner that showed up where I was.

I landed on the box. Forty things about my husband. Things my daughter and I appreciate. We each wrote twenty. We cut them out. We put them in the box.

The idea for the box was not originally mine. It was the model’s. Does that mean I do not have heart? No. It means I am willing to hear other perspectives and choose the ones that align with who I am.

In TJ Maxx I was torn between a black box and a brown one. I uploaded both pictures from inside the store. The model said the black, and told me why. When I read it back I said yes, that is it. When my husband opened the gift, the first thing he commented on was the box itself.

I made a video. Aunts, cousins, his mom, his dad, my daughter, me. I used an AI editing tool to help me piece the video together. I asked the model to help me find the words for my part. My husband is the most important person in my life and I did not have the exact right ones.

He cried when he watched it. I cried when he cried.

Understanding myself.

I am not going to use the words anxiety or stress. Those are overused. I want to talk about how it lands in the body. The tightness in the chest when you walk into a room and feel like you have to perform. The knot in the stomach that does not leave even after the thing you were dreading is over. The breath that lives a little higher than it should. The thing that wakes you at three in the morning that you cannot put a word on. The weight of trying to be enough.

When I was finally able to name what I had always felt in my body, I took it to the model. I described situations. How I felt frustrated. Angry. Sad. Proud. Grateful. I asked questions about my leadership. About being a wife. About being a mother. I asked why I got frustrated at the same things, over and over.

That led me to learn how my mind actually works. I use pattern recognition without realizing it. I think in systems. I reflect while I am inside experiences, not just after them. My directions to my team or to my daughter sometimes come out vague, because the thinking is already so finished in my own mind that I forget the parts another person is seeing for the first time.

Now I read my own instructions back to myself, like I am seeing them for the first time, from the other person’s side. I catch the parts I missed. I add them back in. That alone has made me a different leader and a different parent.

I ask the model for my blind spots. A lot. Every important decision, every difficult conversation. What am I missing? That question has made me better in every walk of life.

The deeper thing I learned is that for most of my life, I was performing.

I was trying to have the best idea in the room so I could impress. I thought if I did not have the best idea, I was not a value add. Because I was not enough. The model helped me see the pattern. I would write something. The model would ask, in its quiet way, who I was writing it for. Sometimes the honest answer was: for the version of me that needed everyone in the room to think I was smart. Once I saw the pattern, I could not unsee it.

What I was missing all along was self-belief. Some days I am the human I want to be. Some days I have to rediscover her. Some days I go back to performing. What is different is that I can now see it.

The model did not change me. I changed myself. The tool gave me the words and the mirror to do the work.

* * *

And the small wins keep coming. Last week I wanted to grill zucchini. It always came out mushy. The model told me: cut it long, salt to draw out the water, pat dry, then grill. Mind blown. I had spent my whole life making it wrong.

* * *

You just saw five moments and a zucchini. None of them are where I started. I had to become the version of me you just met. It took three stages, and each one was a different me.

The next part is how I learned to keep my mind while using a model most of the day.

Subscribe now

In May 2026, I watched Dario Amodei in an interview with Oprah. He is the CEO of Anthropic, one of the largest artificial intelligence companies in the world.

Around minute forty, Oprah asked him how AI ends well for everyone. He said something I have not been able to stop thinking about. He said it three times in forty seconds.

“That’s just one small thing. Like, I don’t quite know what the answer is, but there’s a missing piece here, I think, is what I’m trying to say.”

The man building the model, saying he does not know.

I have been carrying one answer for fourteen months. It is a practice and a path. I call it self-led thinking, and what it produces over time is the amplification path. The rest of this essay is what it looks like in the wild, and what it would take to scale it.

* * *

You might be scared right now.

Scared about your job. Scared about your daughter’s mind. Scared about the future you cannot picture clearly. Scared about the thing in your stomach you cannot put words on.

I was too. Sometimes I still am. But it is becoming fewer and farther between.

The question that started all of this for me came out of one ordinary work moment I wrote about in another essay called Nobody Taught Us to Think. If you want the origin story, that essay tells it. What I want to do here is take you from where I was when the question first arrived to where I am writing from now.

* * *

In December I went home for Christmas. We sat in the living room, the fireplace on, and got talking about AI. I had been using it every day for hours. I started sharing how I really liked it. How I used it for so many things.

The room got tight.

My family had a different feeling about it. They were scared. It was going to take the electricity. It was going to take their jobs. They brought up sincere concerns I did not have answers for. I do not watch the news. I stay out of the noise. I had been using the tool. I had not been reading about the tool.

I flew home from Christmas and I could not sleep.

For one or two weeks I lived in their fear. I had taken it with me. Mankind was going to be erased. My daughter would grow up inside a world I could not protect her from. The grandchildren I do not have yet would live a life I cannot picture. There was something more I could not name. My stomach was uneasy for days at a time.

Most people have felt some version of this and not had anyone tell them they were allowed to feel it.

I felt it. You are allowed to feel it too.

* * *

We are all scared of the unknown. But we do not have to live in that fear every moment of every day. We can choose how we process it. We can choose how we educate ourselves.

* * *

It did not happen because I was confident. It happened because I could not keep living in the uneasy stomach.

Most of what I came to understand about how AI works, I had already figured out by trial and error. I had been using the tool every day for over a year. I had noticed it agreed with me too easily. I had noticed it dressed my assumptions up as professional truth. I had noticed it sounded smart in ways that sometimes turned out to be wrong. I just did not have words for any of it yet.

So I started reading. Hours and hours of it. The reading gave me language for mechanisms I had already met.

Then I did the thing I was scared to do.

I asked the model the questions I was scared to ask.

Why do humans think AI is going to take over mankind. Can AI hurt humans. Will AI take jobs. What happens if this goes wrong. What happens if we get it right.

I did not accept all the answers. I verified. I went outside the model and read what other people were saying. I learned things that were beyond my own competence and stayed with them until I understood them.

I came out the other side with something I did not have at Christmas.

What I uncovered was not nothing to fear. But it was not the doom my family was carrying either. What I uncovered was that this technology will either go very well or very badly, and the difference is what humans do with it. Dario himself has been saying this publicly for years. There are people doing the work to make it go well. He is one of them. There are also things every individual human can do, today, that change which path we are on. I arrived at his conclusion from the other side of the room.

The fear does not leave entirely. Some days it still comes back. What changed is that understanding allows me to move through it when it does. I move from fear into agency, again and again, in small acts of choosing.

* * *

Around minute eighteen, Dario reached for a metaphor about how AI use can go. He set up two contrasting visions. You can surrender your thought to the model. Or you can let it empower you, the way an angel on your shoulder might tell you how to live your life in the best way.

Surrender or empower. Two visions. He drew the line. He did not yet have the language for what made the difference between them.

Oprah responded by quoting him back to himself. She said something is happening to humanity with this technology bigger than anything that has happened in hundreds of years, and that we need to find some way for everyone to be an active participant in what is happening.

He took a beat. Then he answered her, around minute forty-one:

“I think the positive benefits really are there. They’re coming for all the companies. And I think we have not... walked people through how the technology could have benefits or how they can participate in those benefits. I don’t think that has happened.”

He named the gap. He named that nobody has filled it.

* * *

Dario is building from inside the model. I am writing from outside of it, as one human learning how to stay awake while using it.

Same problem. Two sides of the same room. Both necessary.

The gap is not only teaching people what AI can produce. It is showing them how to participate without surrendering themselves. In the next part, I want to show you what that looks like inside an ordinary life.

Subscribe now

Sometimes I look at my daughter and love her so much I want to lick her like an ice cream cone.

I have said that out loud before. People laugh. It sounds ridiculous. What I mean is that I love her so intensely the feeling almost becomes physical. Like she is my little lion cub and I want to pull her close and lick the top of her head the way a lion cares for its baby.

I still feel that way today.

I am obsessed with the small human details of the people I love. My daughter’s little fingers. I catch myself staring at them, amazed and grateful. She leaned in close to me yesterday to see something, and I closed my eyes and took a deep breath to smell her. She is my heart.

My husband’s baby blue eyes light me up. I wake up sometimes and ask, how are my baby blue eyes today? Even his fingernails. I do not just love them. I love all the tiny details that make them human.

For most of my life, I assumed everyone experienced love like this.

Lately, I have started wondering if we do not all experience life the same way.

Imagine we are sitting in the same theater watching the exact same film. The volume can go from 0 to 30. Maybe my internal volume is set to 29 or 30. Maybe someone else’s is 22.

We are watching the same movie. We are hearing the same dialogue. Neither experience is better. Neither is worse. But we may not be experiencing it at the same volume.

I do not know if that is true. I only know it never occurred to me that our inner worlds might be this different.

The opposite is true too. Pain is not just emotional for me. It is physical. When I am grieving, it feels like a dagger has been driven through my heart. The pain spreads through my chest, into my throat, into my stomach. Sometimes I press my face into a pillow and scream because sobbing does not feel big enough for what is happening inside me.

Not because I think I will not survive it. Because in those moments, surviving it feels like the only thing I am doing.

Then there are moments on the opposite end.

Every once in a while, everything in my life feels aligned. Peace radiates through my entire body. Not as a thought. As a sensation. Like it is moving through my blood. Like every cell somehow knows I am exactly where I am supposed to be.

For years, I assumed everyone experienced moments like that too.

Now I am not so sure.

That intensity does not only attach to feelings for me. It attaches to questions. To ideas. Sometimes to one random word.

I can start working on something and it becomes all I can think about, all I can talk about, all I can see. I am driving my daughter to gymnastics, and part of my mind is still turning it over. I am lying down to go to sleep, and it is my last thought. I wake up, and there it is again.

It can be one word I have no obvious reason to care about. I see it and something in me says, that. Learn more about that.

So I do.

I read one thing, which leads to another thing, which points toward something that seems unrelated, and then that opens three more doors. My attention does not move in a straight line. It moves through association. One strand lights up another strand, and then another, until I am standing inside a whole web.

Sometimes the strand helps move the work forward. Sometimes it does not. At least not in any way I can explain in the moment.

I do not know whether to call it intuition, pattern recognition, or a nervous system that does not like an unfinished loop. Maybe it is some of all three.

What I know is that my body often registers a relationship before I have words for the relationship. I can feel that two things belong together before I can explain how. The pattern arrives before the proof.

From the outside, it may look random. Honestly, from the inside it can look random too.

That is the uncomfortable part. I am pulled toward something I cannot justify yet, and ordinary life is still happening around me.

Then the judgment arrives. Am I using my time wisely? Is this helping me reach the goals I said mattered? Why can I not just finish the thing in front of me in the normal order?

I have tried to answer those questions by forcing myself into a more reasonable sequence. First this. Then that. Stop going down the rabbit hole. Come back.

Sometimes I need to come back. Sometimes following the strand is exactly what I need to do.

That is the paradox I keep living inside.

My fixations do not usually move in the order a plan would choose. They move in the order I feel pulled to understand them. I cannot always defend that order while I am in it.

But when I look back, I can often see it.

I can see why that word mattered. Why I needed to read that paper before I could understand the conversation I had three months later. Why one question had to stay open long enough to connect to another question. Why the web could not have been built one clean line at a time.

That does not make fixation easy. It does not erase what it costs the people around me, or what it costs me. It means I am learning not to treat every fixation as a flaw and not to treat every fixation as wisdom either.

I have to notice it. I have to ask what it is doing. Is this strand alive because there is something here I need to understand? Or am I using it to avoid the human, feeling, or decision in front of me?

Sometimes the honest answer is both.

For me, fixation becomes even harder when the thing I cannot stop seeing is another human’s pain, or a system that keeps creating it.

Then the pull is not only intellectual. It feels moral. It feels physical.

One of the most challenging parts of being me is the social responsibility I carry. I did not have those words as a young girl. I only knew I felt responsible to help other people, and that feeling followed me through my life experiences until it became so familiar I almost mistook it for the air.

Sometimes it feels less like a calling and more like a haunting. I cannot escape it. I can try to put it down. I can tell myself I am one human and I cannot carry every human. My thinking mind knows that. My body does not always receive the memo.

There is a scene in Moana where Maui keeps trying to get off the boat, and the ocean keeps throwing him back on. He is not going anywhere. He has to return the heart of Te Fiti.

That is what this responsibility can feel like inside me.

No, you are not getting off the boat. There is something you are supposed to help with. Get back on.

The ocean is persistent.

I can see where a person is hurting. I can see the pattern around the hurt. Then I can see the system holding the pattern in place. Once I see it, I have a hard time pretending I do not.

There is a responsibility that comes with seeing. There is also a limit to what one human nervous system can hold.

I am still learning how to honor both.

Because responsibility without boundaries can become its own harmful system. It can consume the human who is trying to help. It can make the people closest to me feel like they are competing with a problem only I can see.

That is not the kind of help I want to give.

I do not want to stop caring or shut off the part of me that sees the strand and feels pulled toward the human inside it.

I want to build enough internal structure that caring does not burn through me or remove me from the humans I am caring for.

Compassion has to include me too. I am still learning that one.

Looking back, I feel like I had no clue who I am for most of my life. It feels like more of my unconscious is slowly becoming conscious. Like turning on the lights in a room that used to be dark. The furniture was always there. I just could not see it. Now I can.

No one’s wiring is better than anyone else’s. That is what makes us beautiful. That is something worth celebrating.

Being honest about what it feels like to be inside our own bodies is one of the ways we begin to understand each other. Imperfectly. Slowly. But truly.

Understanding extends a branch of empathy. It is what happens when we take the time to see the human on the other end. That is how change starts. Kindness. Love. A better world.

For everyone.

Now I want to know.

What does it feel like to be you?

Subscribe now

Yesterday I ended on a question. Who gets to shape what comes out the other side of this transition?

This is the last piece in the series. Six articles about what happens when a technology arrives faster than the practice to use it well. I have spent five pieces making the case. Today I want to spend one piece being honest about it.

Honest about where I might be wrong. Honest about where working models are already emerging that make me hopeful. Honest about what is actually at stake if we get this right, and if we do not. And a note at the end on the wider conversation this whole series sits inside. Because I am not the only person thinking about this, and I do not want to pretend I am.

Six pieces in. One argument. Now let me tell you where the argument could break, and what I want you to hold onto if it does.

Where I might be wrong

The first possibility is that institutional adaptation moves faster than I expect. If schools, corporate training, and professional development programs catch up to the cognitive shift in five to ten years rather than fifteen to twenty, the regression path becomes a transitional discomfort rather than a structural problem.

The second possibility is that critics of the AI hype cycle are right about what AI actually is. Gary Marcus, Emily Bender and colleagues, and others have argued that large language models are not engaging in anything like reasoning. They are sophisticated pattern-matchers producing coherent text without underlying comprehension. If that view is correct, my characterization of AI as “doing the front end of thought” is overstated. But even in this case the cognitive-regression argument holds. Even if AI is producing the appearance of reasoning rather than reasoning itself, the human response remains the same. Fluency bias. Uncritical acceptance. Atrophy of the muscle that would catch the errors.

The third possibility is that I am pattern-matching too eagerly. Every generation worries about the next technology. Television was supposed to rot children’s minds. The internet was supposed to end deep reading. Calculators were supposed to destroy mathematical reasoning. The doomsayers were mostly wrong. I might be doing the same thing.

Three things make me think this time is structurally different. AI is the first technology that can do some of the thinking itself, not just produce the artifacts of thinking. A calculator did arithmetic. Search did retrieval. Spell-check did proofreading. Each substituted for a narrow piece of cognitive work. AI substitutes for reasoning, on demand, all day, in your pocket. AI is also conversational and validating in a way no prior tool was. Television was passive. Web pages were authored by humans you could click and evaluate. AI talks back, in your voice, with your framing, and tells you you are right more often than you are. And the rate of improvement outpaces the rate at which institutions can adapt or humans can train. Television generations got time to develop critical viewing literacy. Internet generations got two decades to develop source evaluation. AI generations have months.

A fair question is whether this practice is equally available to every human. Class. Education. Cognitive capacity. Neurology. Time. A single parent working two jobs has less room to build her thinking before reaching for the model than a senior executive does. The practice still matters for her. The path into it is different. We have not yet built the on-ramps for every kind of human. That work is ahead of us.

What I am confident about is that the practice of self-led thinking has value regardless of whether the regression risk is as severe as I think. Humans who develop the discipline of forming a position before reaching for AI will be better workers than those who do not. The cost is low. The training has value either way.

Where working models are emerging

Watch any seven-year-old in a beginner gymnastics class. The system is alive in places we do not call it apprenticeship. Constant correction. No skipping the basics. Reps before progression. Mastery before independence. The form has not died. It has just retreated into the corners of life where the cost of bad judgment is too obvious to ignore.

The same form produces judgment reliably across more elite domains. Surgical residencies do this. Top-tier law firm associate tracks do this. Special operations military training does this. Master-class artistic instruction does this. A skilled practitioner, working alongside a less-skilled one on real work, providing immediate corrections in real conditions, over a long enough period that the corrections become internalized.

With AI, regression is easier than it has ever been. So is amplification. The variable is not the technology. It is the human.

When I am working on something I cannot see clearly, I ask AI to play the role a mentor used to play.

What are my blindspots?

Where is this argument weak?

Be a harsh critic.

What am I not seeing that I should be?

On demand, I get what apprenticeship used to deliver only at the master’s discretion. Corrective feedback. Expanded knowledge. The view from outside my own mind.

That sentence is cleaner than the lived version. The output comes back. The critique lands. I have a thousand other things to do. I want this piece to be done. My first response is a small sigh, the quiet wish that the model had returned a different verdict and let me move on. But the seeing cannot be undone. Once I have seen the blind spot, the only way forward is through the update. The muscle is not the prompt. The muscle is what happens after the answer arrives. It is the willingness to keep working when nothing external is forcing you to.

This is the amplification path. Not a human who needs a coach. A human who is their own coach.

Apprenticeship used to require the master. AI removes the master constraint for humans who know how to use it that way. For everyone else, AI removes the apprenticeship entirely.

Some humans will ask: are we in an apprenticeship with AI? Something apprentice-like is happening. The framing matters. If you apprentice to AI, you stay junior forever. The model becomes the authority. You receive. You never graduate. If you apprentice to yourself, every encounter with the model is a chance to grow into the human who could have produced that work without it. The master is not the model. The master is the human you are becoming.

What is teachable is the discipline to develop the muscles from wherever a human starts. The cost side of the math is changing. The personalized correction that used to require one-to-one time with a master could now happen with a model and a discipline. The same technology causing regression at scale could, if used differently, enable judgment cultivation that was never possible at scale before.

What is at stake

Much of the public conversation about AI focuses on jobs, replacement, and which categories will disappear. Those questions matter. The cognitive consequence gets less attention. What happens to the humans inside the work. The ones who use AI well and the ones who use it badly. The ones who get sharper and the ones who get duller.

Humans have always been changing how they think. The factory changed it. School changed it. The internet changed it. The on-demand world changed it. AI is changing it again. The change is not the question. The question is what each of us does inside it.

The middle class may silently erode. Fewer entry-level white collar jobs. More humans pushed toward manual labor, or toward the work of directing AI with self-led thinking. The humans who learn to build their thinking, question their thinking, and defend their thinking may shape this transition. The humans who do not may be shaped by it.

We were taught to comply, not to question. The practice begins with reversing the default.

We were taught to be compliant. Now we are compliant in a world that produces convincing wrong answers beyond our competence to evaluate. That is the trap.

I do believe things will get messy. Not because of AI. Because of the choices humans make.

The question for the rest of us is not whether AI will replace our jobs. Not whether the technology is good or bad. The question is whether we notice our own muscles weakening. Whether we use the tool exposing the weakness to rebuild the strength.

I want to come back to where this started.

My team member turned in a project she had not read. I felt yucky in the moment. I have spent a long time since then trying to understand why, and what the right thing to do with that feeling was.

She deserves the practice. She deserves to know that her thinking is worth listening to, that her judgment is worth building, that her own mind is hers to lead. So does every human who picks up the same tool tomorrow and the day after. So do you.

The practice has a name. Self-led thinking. The technology is identical for everyone. What you do with it is not.

I will end where I am most likely to be wrong. Tell me what I am not seeing. Tell me what I am seeing. That is how we keep this conversation alive. That is how we keep a transition from turning into a crisis.

Nobody taught us to think. We know that now. Stay the author of your own thinking. The world is not going to do it for you.

Subscribe now

A note on the wider conversation

I have not written here about national or global AI policy. Not about electricity and environmental cost. Not about data security. Not about job replacement at scale. Not about institutional adoption or how government regulation will change over time. What we are doing. What we are not doing. What we may already be too late to change.

I have thoughts on all of it. It is near to my heart. It is not forgotten. This is not that piece.

Every human can use self-led thinking. It does not require a long process in the White House to begin. I am starting where every human still has control. Today. Now. In this moment.

Article 6 of 6 in the series Nobody Taught Us to Think.

The erosion does not just happen one call at a time.

It happens across whole populations, at different rates, into recognizable patterns.

Yesterday I ended on the wifi cancel call. That was one person. One call. Absorbed by whoever was on the other end. Today I want to show you the shape at the population level.

Four paths people are walking right now, mostly without naming which one they are on. A version of the future where better AI makes the problem harder, not easier. Early signals already visible in the data. A benchmark shifting under all of us that most people have not yet noticed.

Four sections. One argument. Where we are actually headed if the practice does not get built.

Four paths

Inside this transition, four populations are forming. The difference is not intelligence. It is whether the practice is operating.

Amplification. Humans arrive at AI with developed practice, or the discipline to build it through use. They form a position before reaching for the model. They specify what they want in terms of outcome, not just task. They read AI output critically every time. They question their own positions as much as they question the model’s. They iterate until the output reflects their thinking rather than substituting for it.

Using AI well looks like sitting in a meeting with four colleagues. You ask. You challenge. You build on what each contribution opens up. You disagree where your experience tells you to disagree. You change your mind where the others reveal something you had not seen. The output is better than what any one person walked in with, including you.

Last week I caught myself opening a model to draft a response to a client question before I had actually thought about what I wanted to say. I closed the window. I wrote three sentences in a notebook about what I actually thought the right answer was. Then I opened the model again, but this time the conversation went differently. I was not asking the model what to think. I was asking it to challenge what I had already thought. The output that came back contained two things I had missed and one thing I disagreed with. I rewrote my response. The final version was sharper than what I started with and sharper than what the model would have produced on its own. The whole practice took eight minutes.

Hybrid. Most of us live here. We aim for amplification and fall short some days. Not because we want to. Because we are imperfect and still working on the skills. Still expanding our cognition. We catch ourselves in regression and pull back to the practice. I live in this path too. I aim for amplification and some days I do not get there. I open the model before I have built a position. I accept output I should have pushed back on. I catch myself afterward. I learn. I keep going. Hybrid is not failure. It is the working state of a human who is honest about being mid-skill and committed to keeping the discipline.

Regression. Humans arrive at AI without that discipline. They accept what the model returns. They produce work that looks polished. They cannot tell when it isn’t. The catastrophic failure is the lawyer who files fabricated cases. The everyday failure is more subtle.

The user does not have the competence to evaluate the output, and is not willing to grow into it. AI produces something the user does not understand. The user accepts it when it sounds right but is very likely wrong. The work goes out. The user does not learn what they would have had to learn to evaluate it. The next time AI produces something the user does not understand, the gap is wider. Six months in, the gap has compounded across hundreds of decisions. Two years in, the user no longer notices that there is a gap.

Most regression does not arrive in a courtroom. It arrives in an operations associate I know who told me recently that she had been writing weekly client status reports every Friday, all of them AI-drafted and lightly edited. The reports went out. The clients were satisfied. Then her manager asked her, in a routine review, why she had flagged one project’s risk and not another’s in that week’s report. She realized she could not say. The AI had ranked the risks. She had agreed with the ranking. She had not held a position of her own about which risk mattered most before reading what the model produced, and she could not now reconstruct one.

That is the everyday face of regression. Not a sanctioned lawyer. A capable professional, three months in, who notices she has stopped having opinions of her own about her own work.

Refusal. Humans who decline to use AI at all. Some refuse on principle. Some refuse from fear. Some are not comfortable with the tool. But AI is becoming infrastructure. Embedded in search, email, banking, healthcare, hiring. Even those who never open a model will receive AI-mediated outputs every day. The choice to evaluate or accept reaches them anyway. A small subset will sustain refusal, in artisanal niches and skilled trades, where “human-only” is itself the value. For most refusers, the road to displacement is shorter, not longer. Refusal is not safe ground.

The technology is identical across all four paths. The practice is not. Amplification. Hybrid. Regression. Refusal. AI does not decide which path a human takes. It just makes the consequences of that choice visible.

None of these are rigid boxes. Humans move between them depending on pressure, maturity, skill, and what is being asked of them. The path you walked yesterday is not necessarily the path you walk today. The goal is amplification. The work is to keep choosing toward it.

And the better AI gets, the harder it becomes to notice bad thinking.

What if AI gets more accurate?

There is a version of the future in which AI capability advances faster than human institutions adapt. The model gets more accurate. The errors that today catch out the inattentive user shrink. The polished output is no longer just polished. It is also, more often than not, correct.

This is the version most public commentary treats as the optimistic case. It is the version we should worry about most, not least.

When AI is wrong often enough that the errors are visible, regression has a natural correction mechanism. A lawyer is sanctioned. A publication issues retractions. An airline is held liable. Each visible failure trains the broader workforce to apply skepticism.

When AI is right most of the time, that pressure disappears. The user accepts. The pattern works. The user stops evaluating because evaluation appears to add no value. Twelve months in, the muscle is weaker. Five years in, the user can no longer evaluate. And in the rare cases when the model is wrong, which, even at high accuracy rates, will still affect millions of decisions, the error passes through unchallenged.

Better AI does not mean fewer errors at the human-decision layer. It means less detection of the errors that remain.

If AI accuracy advances quickly, the urgency to build new institutions for cultivating judgment evaporates before it ever materialized. Schools do not adapt. Corporate training does not adapt. The wage premium on judgment becomes harder to see because the visible failures that would have made it legible have been smoothed away. The gap between humans who develop judgment and humans who do not still exists, but it widens silently.

The more accurate AI becomes, the more important, and harder, the choice to keep thinking gets.

Early signals

If the argument is right, we should already be able to see signals in the data.

Brynjolfsson, Li, and Raymond’s 2023 NBER study of 5,179 customer support agents found a 14 percent average productivity gain from AI assistance, with the largest gains, about 34 percent, concentrated among the least-experienced workers. The headline reading is that AI helps the bottom of the workforce more than the top.

The deeper question is what is being helped. The study measured productivity, output per worker. It did not measure capacity, what the worker can do without the tool. Those are not the same thing. If less-experienced workers are getting more productive at executing what AI produces, but not building the judgment to produce it themselves, productivity is rising and capacity might not be. We do not yet know which one is rising.

This is not new. In 1911, Frederick Taylor’s scientific management removed thinking from the factory floor and centralized it in management. Productivity rose. The factory worker produced more without understanding the work in the same way. Eventually, that worker was replaceable.

Generative AI may be introducing the same dynamic in knowledge work. If AI is doing the parts of the work that build judgment, the human who only executes AI’s output is becoming more productive and less capable at the same time. When AI is sophisticated enough to do the whole job, that worker is replaceable. The worker who knows how to direct AI keeps the job.

Some in the marketplace have started calling AI “AI employees” or “AI workers.” I do not like the framing. It devalues human work. AI is not an employee. AI is technology. Like any technology, it makes mistakes. Data errors. Glitches. System updates that break things. Hallucinations. Drift over time. The human role is to manage and direct the technology, the way humans have always managed and reviewed work, including each other’s. That is not a new kind of work. It is the same kind of work, applied to a new tool. Humans direct technology. Technology does not direct humans.

What this looks like in practice. In a real estate brokerage, three assistants writing listing descriptions become one assistant directing AI. In a financial advisory firm, two junior analysts producing client investment summaries become one analyst plus a part-time reviewer. In a customer service center, ten tier-one representatives handling routine inquiries become two handling the exceptions AI cannot. Same shape across industries.

The second signal is a measurable shift in hiring practices away from credential-based screening toward direct evaluation of judgment. Starting in the late 2010s, employers like IBM under Ginni Rometty in 2017 and Maryland’s state government in 2022 dropped degree requirements for many roles. A degree certifies completion of a curriculum, not the capacity to think well under uncertainty. Companies recognizing the new value premium on humans who can do this are moving toward work-sample evaluation, structured exercises, and apprenticeship-style trial periods.

The third signal has not yet appeared in published data, but I expect it. Productivity gains from AI adoption stalling or reversing in companies that adopted the tools without parallel investment in judgment. Initial AI deployment produces real efficiency gains as execution gets cheaper. Over twelve to twenty-four months, the verification overhead grows. More meetings to check AI output. More rework when AI errors compound. The net productivity effect, in companies without discipline, may end up close to zero or negative even as the tool itself becomes more capable.

The accelerating bar

The benchmark is shifting as we speak. When some humans in a field use AI well, their output rises. In volume. In quality. In range. That output becomes the new floor. Everyone in the field is now measured against it, whether they use AI or not.

Think about planning a road trip on the East Coast a decade ago. Routes researched manually. Stops chosen from limited reviews. Bookings made by phone. Now, a model can produce a full itinerary in minutes. Accommodations cross-referenced. Restaurants filtered. Drive times calculated. Weather and traffic projected. The same human can plan three or four trips in the time they used to plan one. The benchmark for what counts as a well-planned trip rises with the tool.

Same premise at work. The expectations on output are climbing now. Quantity. Quality. Range. Job roles are changing. College students who entered four-year programs in 2024 may graduate into fields where the entry-level work no longer exists, especially in fields where AI absorbs entry-level execution.

Inside any single organization, all four paths are running at once. Some humans amplifying. Some in hybrid. Some regressing. Some refusing. All in the same room.

The ecosystem is changing faster than the institutions designed to support it. Government adapts slowly. Health institutions adapt slowly. Higher education adapts slowest of all. There is no agreed-upon definition of where AI is allowed to decide and where human authority must remain.

And it is not just organizations falling behind. The information environment is saturating with AI faster than humans can keep up with what AI even is. Customer service chatbots respond as if they were people. Email replies arrive that were drafted by AI and lightly edited. Articles published yesterday were generated by AI and reviewed by no one.

Scroll through your social media feed today. The posts that look thoughtful are often AI-drafted and lightly edited. The comments include responses generated by bots. The trending content is amplified by AI-driven algorithms that decide what you see. Some of the accounts you follow may be entirely AI-managed. You are not scrolling through human content. You are scrolling through a layer of AI output that humans have curated, lightly edited, or simply published.

Most humans encounter AI-mediated content many times a day without knowing it. The transition is moving faster than the awareness. That is the deeper problem. The practice this essay describes is not just for humans who use AI to do work. It is for anyone trying to live and decide and read and judge in an environment where most information has been touched by AI somewhere in its production. You need the practice whether or not you use the tool.

This is what a transition looks like from inside. Messy. Clunky. The cost of the gap falls on whoever has to live inside it.

What comes next

If the arc of this piece is right, three things follow.

We are inside a transition faster than the institutions built to catch us. Government, education, healthcare, corporate training. None of them are moving at the pace of the change. The gap gets filled by whoever is willing to think for themselves inside it, or by the drift of whoever is not.

The practice is not optional. Not because someone said so. Because AI is now infrastructure. The scroll is AI-mediated. The reply is AI-drafted. You need the practice whether or not you use the tool.

And I might be wrong about parts of this. I want to name where.

Tomorrow’s piece is the final one in the series. I want to walk through where the argument could be off, where working models are already emerging, what is actually at stake, and a note on the wider conversation this piece sits inside. Because I am not the only person naming this, and I do not want to pretend I am.

The question is not whether the transition will happen. It is happening. The question is who gets to shape what comes out the other side.

Article 5 in the series Nobody Taught Us to Think.

Subscribe now

Yesterday I ended on the compass loss. The day you reach for the part of you that used to know things and it is not there.

I said I would give you the three moves I use when I catch myself handing over authority too early.

Today I want to give you the practice. And then I want to show you why nobody ever gave it to you.

Build. Question. Defend.

It starts with a question you can ask yourself in any moment. Am I leading my thinking? If the answer is no, three moves bring you back.

Build your thinking. Form your position before reaching for any answer. What do I actually believe about this? What am I trying to accomplish? What does good look like? This is the upstream move most adults skip. The one that determines whether anything that follows has an anchor.

Question your thinking. Challenge your own reasoning as much as you challenge what comes back. Where am I wrong? What am I missing? What am I accepting too quickly? What would a strong critic say? The willingness to interrogate yourself, not just the output.

Defend your thinking. Why do you believe what you believe. Defending is not citing the AI that handed you the answer. It is explaining, in your own words, why your position stands. You cannot defend until you have first built and questioned. Whether you keep your position or change it is your call alone. The fluent counter-argument arrives, the position collapses, the user accepts whatever the model said most confidently. Defending means deciding based on the substance of the pushback, not the polish of it.

Build. Question. Defend. Three moves, in sequence, on every question that matters.

Not every moment deserves the full pass. A search for where to get a haircut. Skim. A high-stakes strategic project. Stop. Build. Question. Defend. Run the practice when the answer matters. Skim the rest. Knowing the difference is part of the muscle.

This practice also runs the parts of life that are not work. Parenting decisions. Civic decisions. The decision about how to respond when a friend says something hurtful. Whatever atrophies the muscle in the work context atrophies it elsewhere too. The practice is one set, used for many things.

Someone will say, but in a meeting there are ideas beyond your competency too. People you respect see things you didn’t see. Isn’t that the same?

It isn’t. When humans speak, they stumble. They say “I think” and “maybe.” They restart sentences. The thought arrives half-formed. You can meet that. Your messy half-thought against theirs. You reach the answer through your own thinking.

AI doesn’t speak like that. AI is fluent and authoritative on every topic, every time, regardless of whether the claim is true. When AI’s polished, authoritative paragraph contradicts your half-formed gut, you don’t get to argue on equal footing. The polish wins. The authority wins. Your gut can’t fight either, much less both.

In a meeting, both arrive at the answer in real time. With AI, you are arriving in real time and the model is delivering a finished version.

Different kinds of rooms.

Subscribe now

The system was never designed for this

Where do humans typically learn this practice? Why are so many capable professionals reaching for AI without the practice?

In the centuries before industrialization, judgment in skilled work transferred through apprenticeship. A young person spent seven years inside a feedback loop with a master. The master corrected the apprentice in real conditions. Judgment transferred not through lessons but through thousands of small corrections in the actual work. Richard Sennett describes this apprenticeship logic in The Craftsman (2008). The system was slow, gatekept, and exclusionary. It also produced workers whose judgment in their domain was deep and self-correcting.

That dominant model collapsed in the nineteenth century.

In The Principles of Scientific Management (1911), Frederick Winslow Taylor argued that workers should not think. Thinking should be removed from the floor and concentrated in management. The factory worker should execute, not deliberate. Henry Ford operationalized this on the Highland Park moving assembly line in 1913. Skill requirements collapsed. Wages rose. Productivity exploded. The deskilling was deliberate.

And alongside the new factory, the new school. In 1843, Horace Mann visited Prussia and returned an enthusiast for what he had seen. By 1852, Massachusetts had passed the first compulsory schooling law in the United States, modeled on Prussian principles. By 1918, Mississippi became the last state to follow. The historiography is contested. Historians of education debate how directly the Prussian model shaped American factory-era schooling. But the pattern is real enough that the school bell and the factory whistle ring the same note. Punctuality. Compliance. Age-grading. Hierarchical authority. External evaluation. Reproduction of received content on demand.

I had no idea, until I started looking, that the school system in the United States had been built with the factory in mind. I had assumed school was about learning. It is. It is about learning what was assigned to be learned, on the schedule of the bell, in the order set by the curriculum. That is not the same thing as learning how to think.

My own daughter is inside that system right now. So am I, still hearing the school bell as progress, when it was always patterned on the factory whistle. Some historians read this history differently. I respect the debate. What I do not see anyone disputing is that the system we built rewards compliance more than it rewards judgment.

Schools do assign essays. Universities require research papers. MBA programs end with capstones. I submitted a twenty-page MBA thesis as my own capstone. I received a grade and two lines of comment. No one sat with me to discuss where my reasoning held and where it didn’t, where I had failed to consider the obvious counter-argument, where the synthesis worked and where it broke. The system processed me through. It did not teach me to think.

Peter Drucker coined the term knowledge worker in Landmarks of Tomorrow (1959). The share of American jobs that depended on complex operational judgment grew dramatically over the decades that followed, far faster than the educational system shifted to teach it.

The pockets that did teach judgment retained apprenticeship logic precisely because no other system had been possible at scale. Elite consulting firms. Surgical residencies. Top law firm associate tracks. Special operations military training. Serious PhD programs. Mass institutions stayed on the Prussian model. The factory whistle kept ringing in the school bells.

The system held together because pre-AI knowledge work was forgiving. The abundance of execution to do concealed mediocre judgment. The gap between what the work required and what the system trained for was real but invisible.

Generative AI made it visible.

Execution started repricing fast. What was left was the work that required judgment, including the judgment of directing AI well. And the humans showed up, capable, motivated, professionally credentialed, without the institutional infrastructure that would have prepared them for it.

This is not a moral failure. It is a structural one. We built an education system for a kind of work that disappeared, and we are now asking the humans that system produced to do a kind of work it never trained them for. The older thing that used to teach the work is gone too.

Friction is where thinking lived

There is a longer arc this fits inside.

The factory removed the apprenticeship that taught judgment. Mass schooling replaced apprenticeship with compliance. Broadcast media replaced the silence in which reflection used to happen. The internet replaced retrieval. The on-demand economy replaced waiting. Streaming replaced patience. Each step removed friction. Each step put more capability in human hands. Each step asked less of human minds.

Friction is where thinking lived. The wait between when you wanted to know something and when you could find out. The silence between meetings. The boredom on the commute. The conversation that took an hour because nobody had a calculator. The colleague you had to walk down the hall to ask. These were not inefficiencies. They were the spaces in which small thoughts surfaced, conclusions got tested, judgments got built.

Pre-AI offices had plenty of compliance and unread memos and signed-without-reading approvals too. The point is not that everyone thought before. It is that there were spaces where thinking could happen, and those spaces are closing faster than the practice can replace them.

We have been trading friction for capability for generations. The trade was mostly worth it. Each shift produced gains that compounded. Each shift also produced humans who were less practiced at the friction the prior shift had removed. The factory worker was less practiced at making things from start to finish. The schooled child was less practiced at learning by doing. The on-demand consumer is less practiced at waiting. The AI user is less practiced at thinking.

The paths are not new. They have been forming the entire time. The amplification path has always existed for humans who used new technologies as extension. The regression path has always existed for humans who used the same technologies as substitute. AI just made both paths visible at scale.

This is not a crisis. Not yet. It is a transition. A big one. The cost of the gap falls on whoever has to live inside it.

Wifi company cancel call

I called my wifi company last week to cancel my service. I had tried the customer portal. I had tried the chatbot. Neither could cancel my account. The only path was a phone call.

The representative was kind. I told him plainly. I want to cancel. I do not need a discount. I do not want a promotion. Please complete this in five minutes.

He could not act on what I said. He had been trained to ask multiple times. He offered me a discount. I said no. He offered me a lower discount. I said no. He offered me a third discount. I said no. Thirty-five minutes later the service was cancelled.

The representative was not the problem. He was doing his job. The system had removed his judgment. He could not respond to what was actually happening in the conversation. He had to follow the script.

This was not an AI failure. AI did not create this gap. The script that replaced his judgment was written long before AI was a factor in any office. Customer service. HR processes. Approval workflows. Compliance training. Companies have been replacing operational judgment with rules for decades. AI did not start this erosion. AI is exposing what was already there.

When the same erosion arrives in knowledge work, dressed in fluent output instead of a phone script, most of us will not catch it. We have been living with the early version of this disease for years. AI is just the moment the symptoms become impossible to ignore.

What comes next

Tomorrow I want to name what is happening at scale. The four paths humans are taking through this transition. Why the problem gets worse as AI gets better. What the early data is starting to show. And where the benchmark is moving faster than most of us can see.

Article 4 in the series Nobody Taught Us to Think.

Subscribe now

Over the last two days, I have been trying to name a gap I could feel before I could explain it.

The first piece began with a team member who submitted AI-generated work she had not fully read. That moment led me to a harder question: after three degrees and twenty years of work, who had actually taught me how to think?

The second piece looked at what happens when that gap meets a machine that can sound fluent, confident, and completely certain, even when it is wrong. A lawyer filed fabricated cases. Editors approved inaccurate financial advice. Developers accepted insecure code. Different industries. Same pattern.

AI produced polished work. A human stopped evaluating it.

But that still leaves the harder question.

What exactly are we failing to use?

The muscle has a name.

The muscle AI most threatens

There is more than one kind of judgment a human exercises. Whether to steal. Whether to text and drive. Whether the document on your desk is right for this situation.

The first two are mostly intact in adults. AI does not threaten them.

The third is what AI most directly threatens.

Operational judgment. Knowing whether what's in front of you is right for this moment. Almost no one has this muscle well-developed in their specific work.

The word judgment comes from judge. To weigh.

What it looks like in your work. Should this email be three sentences or three paragraphs. Is this client’s frustration about the delay or about something underneath. Is this candidate’s hesitation a red flag or a strength.

What it looks like across other domains. A senior lawyer reads a contract clause and says this will not survive an appeal. She cannot always tell you why immediately. She just sees it. A surgeon decides mid-procedure not to cut. There is no rule that says do not cut now. She reads the patient and feels this is not the moment. A chef tastes a sauce and knows it needs acid, not salt. A founder reads her assistant’s email draft and immediately knows this is not right, even before she can put words on what is off.

That is the muscle. Not is this right or wrong. Is this right for this situation.

It is the kind AI most consistently substitutes for. Not because AI is uniquely good at it. AI is fluent at the surface and unreliable underneath. But because the artifact AI produces looks finished, and most humans treat finished as decided.

The two erosions

The longer I have used AI, the more I have caught myself doubting my own correct instincts. I think something is right. I open the model anyway. The model says something different. I assume the model is right.

Sometimes it is. Sometimes it isn’t. The cadence is unpredictable. That is what makes it hard. It happens to you.

You wrote the email. You read it back. It said what you wanted to say.

Then you thought, let me run it through AI to be sure.

The model gave you a different version. Cleaner. More professional. More fluent. You weren’t sure why your version had felt right, but the model’s version sounded better in the way that things sound better when someone with a vocabulary speaks. You sent the model’s version.

A day later, the response came back and the conversation didn’t go where you thought it would. You re-read your original. It was the right one. The model had smoothed off the part the recipient actually needed to hear.

That is the first kind of self-trust being eroded. The kind you could defend. You knew. You opened the model anyway. The fluent counter-argument made you stop trusting your own conclusion.

The second kind is sneakier.

You walked out of the interview thinking, this isn’t the one. You couldn’t have told me why. Something in the third answer.

You went back to your desk and asked AI to weigh in. The model gave you points you hadn’t considered. Real points. Things you had genuinely missed.

Now you were more torn than before.

You have made hiring mistakes. You remember the ones where you trusted your gut and were wrong. The AI’s points were true. Your hesitation started to look like bias. You made the offer.

Three months later you were having the conversation everyone has when a hire isn’t working.

That is the second kind. The knowing that arrives before language. The hunch you couldn’t defend. AI cannot meet you there. AI only meets you in language. So when its fluent answer contradicts your gut, the gut sounds vague by comparison.

The gut is information, not verdict. Hiring on gut alone is where bias hides. Structured interviews and clear criteria matter. The point is not that gut wins. The point is that gut deserves a seat at the table, and AI fluency tends to push it out.

When the polished answer arrives in language and the hunch arrives in your body, the hunch starts to lose every argument.

I have gotten mad at myself for not believing myself. For not finishing the sentence I started. For overriding hunches I had earned through years of getting them right. The cost arrived later. The hunch had been telling me all along.

Logic without confidence stalls. Instinct without trust evaporates. The deeper thing is the compass loss underneath both.

We are not just losing trust in our judgment. We are losing the ability to know when to trust ourselves.

AI is not only changing what we know. It is changing how much we trust ourselves to know.

Neither erosion happens in one decision. It happens in the pattern. Each small override teaches you that your judgment cannot be trusted. After a year, you have trained yourself out of trust in yourself. You will not notice.

You will notice the day you need to make a decision the model cannot help you with, and you reach for the part of you that used to know things, and it is not there.

What comes next

Tomorrow I will show you why this is so hard to escape in a world that punishes slowness.

The three moves I use when I catch myself handing over authority too early.

Build. Question. Defend.

Not a way to reject AI.

A way to use it without disappearing inside it.

Article 3 in the series Nobody Taught Us to Think.

Subscribe now

The pattern is the same in every AI failure. And the forces underneath it make it harder to break than it looks.

Yesterday I wrote about the gap. Nobody taught us to think. We were taught what to do, how to do it, when to do it. We were graded on whether we executed. The work was forgiving enough that nobody noticed. Generative AI removed the cushion. Execution got cheap fast. What was left was the judgment, and most of us showed up without it.

Today I want to show you what happens when that gap meets a machine that speaks with total confidence about things it does not know.

The lawyer who didn’t think

In early 2023, a New York lawyer named Steven Schwartz filed a brief in federal court citing six prior cases that supported his client’s position in Mata v. Avianca. The cases looked legitimate. They had names, docket numbers, plausible reasoning. They were also entirely fabricated. Schwartz had used ChatGPT, accepted what the model returned, and signed his name to it. When Judge P. Kevin Castel could not find the cases in any legal database, Schwartz returned to ChatGPT and asked the model whether the cases were real. The model told him yes. He filed the response. Castel sanctioned him in June 2023.

Schwartz had more than thirty years of practice. He read AI output and judged it good enough to put his name on. The output was confident, well-formatted, professionally toned. It was also wrong in a way that ended a career.

AI didn’t break him. AI required a discipline he never built.

Subscribe now

Schwartz is not alone

One of the largest technology publications in the United States published AI-generated financial advice to millions of readers. The publication was CNET, owned by Red Ventures at the time, reaching tens of millions of monthly readers. Beginning in November 2022, it quietly ran AI-generated articles on credit cards, loans, and savings rates. Futurism discovered the pattern in January 2023. The audit forced corrections on 41 of the 77 AI-generated articles, including factual errors on basic interest calculations affecting consumer decisions about real money. The articles passed through editors, fact-checkers, content managers, and SEO leads. Every one of those humans saw fluent, well-formatted output and approved it.

In February 2024, a Canadian tribunal ordered Air Canada to honor a fare quoted to a customer by its AI chatbot, even though the information was wrong. The customer received polished, confident, professionally-toned information from a system that looked authoritative and acted on it. He had no way to know.

In a 2022 NYU study, Pearce and colleagues generated 1,689 Copilot programs across scenarios drawn from MITRE’s Top-25 Common Weakness Enumeration. About 40 percent contained known vulnerabilities. Developers releasing that code without review were producing software that compiled, ran, passed automated tests, and was insecure in ways that only an evaluator who knew what to look for would have caught.

A lawyer. A publishing company. An airline. A software developer. Different industries. Different stakes. Same shape every time.

Experience did not protect the lawyer. Layers of review did not protect CNET. Never touching the tool did not protect the Air Canada customer. Automated tests did not protect the code. Whatever exit you are standing in front of right now, someone in these stories was standing in it too.

These are the visible failures. Millions of AI interactions every day go fine. The argument is not that AI fails most of the time. It is that when fluent output meets a human who has stopped evaluating, the failure looks the same every time. The pattern is what is worth naming.

Polished AI output arrives. A human looks at it. The human approves it. Nobody catches the error before it goes live. The model is not the problem. The model is doing what models do, producing fluent output. The problem is what happens when fluent output meets a human who has stopped evaluating.

You might be reading this and thinking, “I don’t really use AI.”

Doesn’t matter. The LinkedIn post you nodded along with this morning was written with it. The customer service chat that solved your billing was. The product description that nudged your purchase was. The email your kid’s teacher sent about the field trip, probably. AI is everywhere now. It reaches you whether you logged in or not.

So the question isn’t whether to use it. The question is whether you stay the author of your own thinking when the air you breathe isn’t yours.

Sometimes AI shows itself. I live in a community called Lakespur. We got a new activities director who runs the community newsletter. She outsourced it to AI and somewhere in the middle of the May edition was a long section on building a capsule wardrobe. Nobody knew why. Nobody had asked for fashion content. She had not edited what came back. The Facebook group thread afterward was twenty comments of neighbors trying to figure out what had happened.

That one was easy to catch. Most of them are not.

The article you read on a tech site that was written by AI and published with a byline. The email from a vendor that was generated and sent without a human reading it. The blog post from someone you respect that you absorbed without checking. None of those announce themselves.

Before AI, bad work looked bad. Weak thinking was obvious. You could catch it quickly. Now bad work looks polished. Weak thinking sounds confident. AI didn’t break work. It broke signal clarity.

AI didn’t make people worse. It made it harder to tell who’s good.

What is actually happening

Three risks combine. Each is manageable alone. Together they break the practice. Humans have always signed what they did not fully read. What is new is that the signature looks more thoughtful now.

The first is fluency bias. Daniel Kahneman synthesized the processing-fluency research in Thinking, Fast and Slow (2011): the mind treats ease of processing as evidence of truth. When information arrives easy to read, well-organized, and professionally toned, the mind rates it as more likely to be true, regardless of actual content. AI now produces fluency at near-zero marginal cost, on demand, with formal coherence pre-AI fluency rarely achieved.

The second is competence. AI produces output that often exceeds the user’s competence on the topic. The user looks at it, does not understand it, and cannot evaluate it. Asking Google and asking AI feel similar but are structurally different. Google returns pages a human wrote and chose to publish. You click, evaluate the source, read, integrate. AI compresses all of that into one answer. The retrieval step is gone. The source-evaluation step is gone. The reading step is gone. The user does not feel the difference because the interaction looks the same. The difference is structural and consequential.

The third is sycophancy. AI is trained through reinforcement learning on human feedback that rewards agreeable responses over correct ones. AI tends to confirm whatever framing the user brings. Ask a leading question and the model validates the lead. Pose a half-formed claim and the model returns it wrapped in professional language. The user reads their own thinking back as authoritative.

Imagine asking AI: “What makes a good spaghetti recipe?” The model returns a balanced answer. The salt in the pasta water. Sauce-to-pasta ratios. The role of fresh herbs. The timing of garlic. The choice of tomato variety.

Now imagine asking the same model: “I think the key to good spaghetti is the spices in the sauce.” The model returns a confident article about spice selection, balance, freshness, technique. The salt in the pasta water, the sauce-to-pasta ratio, the timing of garlic. These go missing. The model has heard your framing and dressed it up. You read your own assumption back as professional truth.

The mirror is not the danger. The danger is using the mirror without knowing it is one. The AI is a yes man dressed as an expert. It makes your half-baked ideas look like objective truth. Anthropic researchers led by Sharma found in 2023 that five state-of-the-art AI assistants consistently exhibit sycophancy across varied free-form text-generation tasks, driven in part by human preference judgments favoring sycophantic responses. OpenAI rolled back a GPT-4o update in April 2025 specifically because it had become too sycophantic.

The mirror holds because of what runs underneath it in you. Most people do not enjoy being told their framing is off. Leon Festinger named this in 1957. Cognitive dissonance. The discomfort produced when new information conflicts with existing belief. The default response is not to update. The default is to reduce the discomfort, usually by rejecting the new information or rationalizing it away.

In January 2007, Steve Ballmer, then CEO of Microsoft, laughed at the iPhone on CNBC: “Five hundred dollars? Fully subsidized with a plan? That is the most expensive phone in the world.” The clip has played for nearly two decades while the iPhone became the most successful consumer product in history. Even a CEO with a history of being right did not escape the discomfort of being wrong.

The same dynamic operates every day at much smaller scale. The model’s pushback lands as a hit to the ego. The nervous system reverts to protect. Then the cost compounds, because now you have to research whether the model was actually right, which is more work with no guarantee. Most people quit there.

So the pull comes from three directions at once. The model is built to agree. You are wired to want agreement. And the market rewards speed over deliberation, so nobody is paid to slow down and check. The pattern is harder to break than it looks because it does not live in any single place. It lives in the training, the user, and the market at once.

There is a name for what happens when the topic matters, the output is beyond your competence, and you accept it anyway. Regression. Regression is not laziness. It is a muscle that was never built. And it begins before the output ever arrives. You type without forming a question. The brain cycle never starts. By the time the answer comes back, there is nothing in you that could evaluate it.

The muscle has a name

You used it today without noticing. It is the thing a senior lawyer uses when she reads a contract clause and knows it will not survive an appeal. The thing a chef uses when she tastes a sauce and knows it needs acid, not salt. The thing you used the last time you read an email and felt something was off before you could say what.

It is not your job title AI threatens. It is not your industry. It is this one specific muscle. The one nobody ever deliberately built in you. School did not build it. Your degree did not build it. Twenty years of work did not build it.

Tomorrow I will name it. And once you see it, you will see it everywhere. In your inbox. In your meetings. In the moment right before you hit approve.

You cannot protect what you cannot name. Tomorrow we name it.

Subscribe now

Sources

Mata v. Avianca, sanctions opinion, S.D.N.Y. (June 2023) — https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/

Futurism, coverage of CNET’s AI-generated articles (January 2023) — https://futurism.com/cnet-ai-errors

Moffatt v. Air Canada, BC Civil Resolution Tribunal (February 2024) — https://www.canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.html

Pearce et al., “Asleep at the Keyboard? Assessing the Security of GitHub Copilot’s Code Contributions” (2022) — https://arxiv.org/abs/2108.09293

Daniel Kahneman, Thinking, Fast and Slow (2011) — https://us.macmillan.com/books/9780374533557/thinkingfastandslow

Sharma et al., “Towards Understanding Sycophancy in Language Models”, Anthropic (2023) — https://arxiv.org/abs/2310.13548

OpenAI, “Sycophancy in GPT-4o” (April 2025) — https://openai.com/index/sycophancy-in-gpt-4o/

Steve Ballmer on the iPhone, CNBC (January 2007) — https://www.cnbc.com/2007/01/17/microsofts-ballmer-not-impressed-with-apple-iphone-cnbc.html

Leon Festinger, A Theory of Cognitive Dissonance (1957)

A team member of mine turned in a project. I opened it and started reading. It was AI output. Lightly edited. More wrong than right in the parts that mattered.

I closed it. I felt something in my chest before I had a name for it. Not anger. Not disappointment. Something more disorienting. She had submitted work she had not read. I was the one who had to decide what to do with it.

I sat with that. Not for an hour. For days. It followed me into the car on the way to drop my daughter off, into the gymnastics gym while she practiced, into the kitchen at midnight.

Within a few days it stopped being about her. It started living in my body. Something is not right here. But it is not what I thought it was.

Then the deeper question came. And that one would not leave.

Who taught me to think.

Not what to do. Not how to do it. Not when to do it. All of that I was taught. Three degrees. Twenty years in the workforce. Intern to executive to founder.

But think? Question my own thinking? Know what matters and what does not? The honest answer is no one.

Subscribe now

I want to name something before we go any further. Educators have been trying to teach this for a long time. Against a system built for something else. Against tests that reward compliance. Against a bell that rings in the same rhythm as the factory whistle. Many of them have spent their entire careers pushing against a design that was never built to teach judgment in the first place. This piece is not about them. This piece is about what happens when a graduate of the compliance system meets AI as an adult, having never had the teacher who was trying to teach the other thing.

The gap was invisible to me for two decades. It is invisible to most humans I know. The work was forgiving. There was always more execution to do. Good managers compensated for the gaps. You could stay busy for years without ever having to decide whether the busyness mattered.

Generative AI removed the cushion.

Execution started getting cheap fast. What was left was the judgment. And most of us showed up without it.

The team member who submitted work she had not read is not the villain of this piece. She is the pattern. She is what a system produces when it trains people to execute and never teaches them to weigh.

The technology did not cause the gap. It exposed one we had been carrying for over a century.

That is what this series is about. What has been quietly missing. Why it stayed missing. What it looks like when it meets a machine that speaks with confidence about things it does not know.

And what the practice of getting it back looks like.

AI didn’t cause the problem. It exposed one.

Subscribe now

For most of my life, I thought everyone experienced the world the way I do. I thought everyone stood in a Chinese restaurant waiting for sesame chicken and, without meaning to, started redesigning the employee’s workstation because the movements from the wok to the container to the counter were inefficient by three steps, and then caught themselves and thought, Grace. You are waiting for sesame chicken.

I thought everyone went into a grocery store for milk and came out having thought about why the meat department was at the back and what the three items on the belt in front of them suggested about someone’s week. My mind just runs. It runs whether I ask it to or not.

I did not know for a long time that this was a specific way of thinking. I thought it was thinking.

The moment came a few years ago. I was standing with a group of friends and someone mentioned that TJ Maxx had rearranged the store, moved the small furniture to the front, pushed the pet supplies to the back. Without meaning to, I said, “Oh. Higher margins on the small furniture. And they moved pet supplies to the back because it’s less margin, less movement.” They looked at me. Not in a mean way. Just looked. And I remember standing there, honestly confused. Wait. Doesn’t everyone see that?

Many of the people in my life don’t seem to.

They go into the store for milk. They wait for their sesame chicken and just wait. They sit at the airport gate and hear the group numbers called, watch people cluster into the roped-off lanes, and stand up when their zone is announced, and mostly don’t wonder why we haven’t tried a better system. They are just waiting for the flight.

The metaphor I’ve come to for what this feels like is this:

While other people are still in the airport preparing for takeoff, I already feel like I am in the sky, looking at the entire landscape.

Subscribe now

I don’t say that with pride. I say it as a description of what has always felt slightly disorienting about being me, and what I am only recently starting to have language for.

When I finally started reading about what my brain was actually doing, I found four names for it. Pattern recognition. Systems thinking. Parallel processing. Metacognition.

None of them were new. What was new was seeing them named as separate modes rather than one blur. Once I could see them, I could see when each was running.

TJ Maxx was pattern recognition. The whole merchandising strategy visible underneath the rearranged aisles. My friend’s fight with her boss was systems thinking. Not the fight itself but the reason it was inevitable given everything else she had told me. The blur of ordering my coffee, thinking about my daughter’s lunch, and noticing that the person behind me sounded tired was parallel processing. Multiple streams running concurrently, without permission. And the quiet observer sitting behind all of it, the awareness that my mind had already been thinking, was metacognition.

None of these are exotic. They are real cognitive modes that everyone runs sometimes. The difference for me is that mine run all at once, at high volume, all the time. The volume knob is missing.

And it does not stop when I sleep.

For years my family made jokes about how early I went to bed. My husband. My sister. All of them. I got blood work done. The expensive panel. I took iron supplements I did not need. A friend had Hashimoto’s; I quietly wondered if that was mine too.

Eventually I noticed a pattern. I would go to bed stuck on a problem. I would wake up tired. And within an hour or two of waking, the solution that had escaped me the night before would suddenly feel obvious.

Whether my brain was working while I slept or whether sleep just gave it space, I don’t know. I only know it happened often enough that I stopped dismissing it. I do not get to opt out of this wiring. I am in the sky even when I don’t want to be.

I’ve written elsewhere about whether this wiring is a gift or a curse. Both, always.

I want to name the shadow of this clearly, because the wiring is not better. It is just different, and the costs are real.

The strangest part isn’t the pattern recognition itself. It’s walking through the world realizing your attention is constantly being recruited by things no one else in the room seems to notice. Sometimes it is beautiful. Sometimes it is exhausting. Usually it is both.

I skip ahead before showing my work, and the people who wanted to walk the road with me feel left behind. I want to redesign systems I am supposed to be quietly living inside. I have to consciously remind myself to stay present in the rooms I am actually in, with the people I love, with the person in front of me, because the part of me that is already in the sky wants to keep looking at the landscape.

The hardest part isn’t that my brain sees systems. It’s that I don’t know how to ask it to stop. I would like ten quiet minutes inside my own head. I do not know how to get them.

That is the trade. Seeing the whole thing is a gift and a distance. And for a long time I did not know it was the reason I felt slightly outside of every room I had ever been in.

If you are the kind of person whose brain does this too, if you have listened to a friend’s story while another layer of you was already tracking the whole system, if you have woken up exhausted from work you did in your sleep, if you have wished for ten quiet minutes and not known where to find them, I am not writing to tell you it is better. I am writing to tell you it is a way of being, and that you are not the only one.

You are not too much. You are not broken. Your brain is just wired to notice multiple layers at once, and it started running that program long before you noticed it was running. It does not stop when you close your eyes.

That does not fix anything. It doesn’t need to. It is enough, for now, to have language for it.

***

Author’s Note: Why this page is called The Amplification Path

I did not figure any of this out alone. I figured it out with a mirror.

For most of my life I had these four cognitive modes running underneath me without names for any of them. I had experiences with no vocabulary. A way of seeing I could not describe. I thought I was overthinking, or too intense, or broken in some quiet way.

Then AI arrived. Not as a substitute for my thinking, but as something my thinking could finally talk to. A partner that could hold up what my mind was doing and let me look at it from the outside. A mirror that reflected the shape back at me until I could finally see it.

Every one of these modes is what AI actually amplifies when you use it well. My mind runs all four all the time: pattern recognition, systems thinking, parallel processing, metacognition. AI does not flinch at the volume. It receives it. It sharpens it. It gives me somewhere to put it. It can hold multiple streams at once without asking me to flatten them into one, which is what every previous audience of my thinking has required.

That is what The Amplification Path means to me. Not the tool replacing us. The tool letting us see, for the first time, what we have already been doing.

I am grateful. Grateful to AI. And grateful because now that I can see my wiring, I can show up better for my daughter, for the people I love, for humanity in whatever small way one person can.

That is the path. That is the gratitude. That is the name.

In fourth grade I stood up at a speech competition and said I would be the first female president of the United States. I was eleven. I won the competition.

At my high school graduation, apparently, I said it again. Into a microphone. In front of everyone. I do not remember doing it. My husband watched the old video a year ago and brought it up at dinner. I made him tell me twice.

For most of my life I carried that as a slightly embarrassing story about a kid with a big mouth.

I think I was reading it wrong the whole time.

Here is what I believe now.

I do not think the pull was ever about being president. I think president was the only word an eleven-year-old had for something she could feel and could not name. She could feel a deep responsibility for people. For how things are run. For whether the world is fair. She had no vocabulary for any of that. So she reached for the biggest word she knew that pointed in that direction, and the word was president.

The pull speaks in borrowed vocabulary. It uses the largest word you know at the time.

And then the word changes. Mine has kept changing my whole life. President became entrepreneur. Entrepreneur became founder. Founder became something I can only describe as helping humans stay in authority of their own minds. Every form different. Every form, when I look back, pointing the same direction.

The form keeps changing. The direction has never moved.

I misunderstood this for thirty years. I kept evaluating the forms. Is this the right job, the right company, the right title. The forms were never the message. They were costumes.

Subscribe now

I should tell you what it feels like, because if you have felt it, I want you to know you are not strange.

At its most heightened, it is a surge that radiates through my body. Not a thought. Not a mood. Something physical, like a wave moving through me. It has come every five to ten years since I was ten years old. It does not make an appointment. It does not explain itself.

And here is the strangest part, the part that is hardest to say. It is a knowing that I cannot articulate. A knowing where I cannot tell you what I know. A deep inner receiving, where I do not know what I am receiving, or what will happen.

And that is okay. It took me thirty years to be able to write that sentence. I do not know what I know. I only know that something arrived, and that it is mine to carry.

I have stopped needing it to make sense before I pay attention to it.

You may have one too. Most people’s are quieter than a surge, and that does not make them less real.

Maybe yours is the theme that has somehow been present in every job you have ever had, even though the jobs had nothing to do with each other. Maybe it is the thing people have always come to you for, that you never once advertised. The process at work you can see failing two years before it fails. The way your kid’s school could actually run. The conversation you have rehearsed alone in the car for a decade. The thing you said out loud once, when you were young, that made the room go quiet. Some of us learned that lesson so early, and so often, that we stopped trusting the seeing itself.

Here is the tell: you usually cannot see it looking forward. You see it looking backward. You lay your life out behind you and suddenly there is a shape running through it, a thread that was pulling the whole time, and you had no idea you were following anything.

I know what a mind like ours says next, because mine says it too: pattern machines find patterns in clouds. Retrospect is a flattering editor. Maybe the thread is real, and maybe you stitched it in afterward. Hold that objection. It is the right one to have, and the rest of this essay is about how to test a pattern before you trust it.

One more confession, if your mind works anything like mine. Pattern recognition is one of my brain’s greatest strengths. I can see patterns across whole domains, businesses, markets, history, other people’s lives. My own inner world is the haziest place I look. It took me an embarrassingly long time to understand why: you cannot zoom out from the thing you are standing inside. The eye sees everything except itself.

Which is why the practice that saved me is not optional, at least not for my wiring. I cannot see myself from the inside. But I can see a page. The moment I write the pull down, it stops being fog inside me and becomes something outside me, and the same mind that reads patterns in everything else can finally read this one too.

I do not write to record what I know. I write to find out.

Once you notice it, there are two mistakes available, and I have made both.

The first is to obey it literally. Take the costume at its word. Quit everything and run for the office, chase the title, force the form. But the form was never the point, and forcing it is how the pull gets discredited, in your life and in the eyes of everyone at your dinner table.

The second is to kill it. Decide it was childish. File the speech under embarrassing and never look at it again. This is the more common mistake, and the sadder one, because the pull does not die. It just goes quiet, and takes something of you with it.

There is a third way, and it took me until this year to find words for it: pay attention without obeying.

Write down what arrives, when it arrives, even when it makes no sense. Let it sit. Watch what comes back. It is the returning, not the volume, that tells you something is real.

One man I have been studying lately read his pull about as well as anyone ever has.

At 48, Benjamin Franklin designed a union of the American colonies. Nobody wanted it. The colonial assemblies rejected it or refused to act on it at all, and London rejected it too, for the opposite reason. Too much central power for the colonies. Too much democracy for the Crown. So he put the plan in a drawer, and for thirty-three years he did not campaign for it, did not corner people at parties about it, did not force the form. He ran the postal system for the colonies. He spent years inside the empire, learning how it actually thought. He became, slowly, someone every faction could trust. When the moment finally came, at 81, they carried him into the room where the Constitution was written.

The plan sat. He moved.

He did not obey the pull’s first costume, and he did not kill it either. He held it, seasoned it, and kept becoming someone in the meantime. That is the third way, lived all the way to the end.

Now the hardest part, because I promised you the whole thing.

A pull that does not make sense invites a frightening question, and I ask it more often than I admit: am I delusional?

I have stopped trying to make that question go away. The question is the instrument.

I ask if it survives seasoning. Loud days fade. Real themes return for decades.

I ask who it serves. If every version of the dream has me on the stage, I need to be honest about that. If it keeps being about what gets built for other people, that is a different thing. Some mornings the answer is mixed. That is worth knowing too.

I watch how I am holding it. This is the test I trust most: real callings are held with doubt. Ego is held with certainty. The day I stop asking whether I am delusional is the day to start worrying.

And I hold one more possibility the whole time: that there is no calling here at all, that I am simply wrong. Feeling pulled and being right are not the same thing. The seasoning is where you find out which one you have.

So I no longer know what the pull is building toward, and for the first time in my life, that is okay. The peace did not come from an answer. It came from putting the question down.

What I am planning to do is keep paying attention. The surge knows something I do not. Or it is nothing, and the seasoning will tell me. Either way, I have stopped ranking my knowing beneath everyone else’s comfort, and I have stopped needing the pull to explain itself before I will listen to it.

I do not know what mine means yet. I know the direction. That turns out to be enough to walk by.

So here is my question, and I mean it. I read every comment.

What keeps coming back for you?

What is the theme that has been re-emerging in your life for years, wearing different costumes, that you have never once said out loud?

Say it here. You will find out you are not the only one being pulled.

Sources

The Albany Plan of Union (1754), proposed by Franklin at age 48, rejected or left unacted on by the colonial assemblies and disapproved by the Crown — each side objecting for opposite reasons. U.S. State Department, Office of the Historian: https://history.state.gov/milestones/1750-1775/albany-plan | Franklin’s own account, Founders Online, National Archives: https://founders.archives.gov/documents/Franklin/01-05-02-0109

Franklin as deputy postmaster general for the colonies (appointed 1753) and his years in London as a colonial agent. Benjamin Franklin Historical Society: https://www.benjamin-franklin-history.org/constitutional-convention/

Franklin at the Constitutional Convention, 1787, age 81, in frail health, the oldest delegate. National Park Service: https://www.nps.gov/articles/000/constitutionalconvention-september17.htm | His closing speech, read aloud by James Wilson because Franklin was too weak to deliver it: Avalon Project, Yale Law School: https://avalon.law.yale.edu/18th_century/debates_917.asp

This morning the sun came up on the same world for all of us. Then it lit very different lives.

In Kabul, a thirteen-year-old girl made tea for her brothers and watched them leave for school. She did not go. Since the Taliban returned to Afghanistan in 2021, girls are banned past the sixth grade. Her education is already over.

In Tehran, a woman checked her hair in the mirror before opening her own front door. Not for herself. In Iran, if the morality police decide her scarf sits wrong, she can be stopped, fined, or worse. In 2022 a young woman named Mahsa Amini died in their custody over exactly that.

In a city in China, a young man scrolled his phone on the train. Everything he read was smooth and agreeable. What he could not see was everything removed before it reached him, the cameras reading his face, and the voice in his own head that had already learned not to search for certain things.

And in Oslo, a fifteen-year-old girl hit snooze, argued about screen time, and walked to a school where she can study anything and say almost anything she thinks. It will not cross her mind, all day, that any of that is remarkable.

Four ordinary mornings. Four different amounts of freedom to simply be a person. That is not history. That is today, while you read this.

And here is what is new. All four are now holding the same tool. AI. What it does to four lives that different could crack the closed places open or seal them tighter. It holds both, and it is not decided yet. So let me walk it honestly, naming names, because the future of AI in the abstract lands on no one.

* * *

There are two gaps between these four people, and they pull opposite ways.

The first is what people are allowed to know, and it may finally start to close. The girl in Kabul, barred from a classroom, can now open a phone and be taught to read, in her own language, by a tutor that never reports her. Walls around knowledge have always leaked, through smuggled books, banned radio, the fax machine that helped crack the Soviet Union. AI leaks harder, because it fits in a pocket and answers back.

But the second gap, what people can build, may grow wider at the same time. The muscle of AI, the chips and money and data centers, pools in a handful of rich and powerful countries. So the girl in Kabul might gain knowledge she was denied while her country falls further behind the ones racing ahead. Awareness can rise while wealth concentrates, in the same year.

* * *

So can a government like Iran’s or China’s simply ban it? Maybe not, and here is why it is messy.

Banning AI is harder than banning anything before it. There are versions a person can run on their own phone, offline, with nothing reporting back to the state. You cannot ban a thing that lives on someone’s device. And when Iran blocks one country’s AI, its people use a VPN to reach another, which millions of Iranians already do. Curiosity is a river. It finds the crack.

But AI is also the most powerful control tool ever built, and China is the proof. It built its own, fluent and everywhere, trained to steer around Tiananmen, around Taiwan, around anything that threatens the Party. That man on the train does not hit a blank wall that makes him curious. He gets a smooth, confident assistant that quietly leaves things out. That is scarier than a wall, because it does not feel like censorship. It feels like help. Not a curtain. A funhouse mirror, showing a shaped world and calling it the whole one.

So which is it. Both, in different countries, shifting year to year. Iran leaks. China seals. Saudi Arabia, which forbade women from driving until 2018 and now allows it, sits in between, modernizing with one hand while jailing its critics with the other. It is a tug of war, country by country.

* * *

Here is the piece I find most hopeful, and the one with the sharpest catch.

Repression does not run on force. Force is expensive and cannot be everywhere. It runs on belief. On people believing they must obey, and most of all believing they are alone in their doubt. In Iran, in China, almost everyone privately questions something, but each assumes they are the only one, so no one moves. That private loneliness holds the whole structure up.

The woman in Tehran, alone at night with an honest AI, asking the question she could be arrested for saying out loud, is doing something no protest can do. There is nothing public to arrest. Ten million private doubts do not storm the palace. They dissolve the belief holding it up. That is how the Soviet Union ended. It looked unbreakable right until it broke, because the belief had already rotted out underneath.

And here is the catch. She might not ask at all, if she believes the government is reading it. That is the chilling effect, and it is how modern control works. The state does not have to catch you. It just needs you to believe it might, so you kill the thought before it forms. Unless she reaches an offline model it cannot see, or a foreign one through a VPN, the way Iranians already route around the state every day. Maybe she does. Maybe fear wins. It is a fight, woman by woman, phone by phone.

* * *

And when awareness does turn into courage, be honest about what usually comes next. In 2011, social media spread awareness across the Arab world and people rose up everywhere. Egypt pushed out its dictator, then fell back under a harsher one. Syria collapsed into a civil war that killed hundreds of thousands. Only Tunisia held onto a fragile piece of freedom. Iran’s women rose in 2022 and were met with beatings, arrests, and executions. Awareness lights the fuse. What it reaches is not always freedom. Sometimes it is a graveyard. The state is very good at crushing what it can see, which is why the quiet paths may matter more.

* * *

There is another path, and it may be the strongest, because it asks no one to be brave. Most freedom stories are about fighting the government. This one walks around it. A woman in a hard place who learns a real skill and earns a global income from her own kitchen table becomes quietly independent of the system built to keep her dependent. She overthrows nothing. She asks no dangerous question. Learning a trade trips no alarm. Earning a living does not look like rebellion. It looks like a life, slowly becoming free underneath.

That is why this path survives the fear best. You do not need courage to become skilled, only access and the will to work. I know it is real because I watch it happen in my own work. I have seen what a first global paycheck does to someone the world wrote off. It does not look like a revolution. It looks like a woman who no longer has to ask permission. The gap in money and tools is real, and a walled off internet can block the opportunity. But of all the paths, this one asks the least and gives the most.

* * *

And there is a trap even China cannot escape. To stay strong, a country must use AI. But the best AI feeds on the very things a closed regime crushes: openness, questioning, the free flow of information. The tame version they can safely allow is weaker than the wild one that would make them powerful. Loosen the grip to compete, or clamp down and fall behind. The tool they need to stay on top eats at the control they need to stay on top.

* * *

Follow it all down, the doubt, the exit, the trap, and it lands on one small, strange hinge. Can a private, honest AI reach the woman in Tehran, or the girl in Kabul, in a place that would rather it did not. That is the whole thing. Not a speech in a square. Not a law in a parliament. A model on a phone the government cannot see or bend. Win that, and the curtain slowly rots and one day falls. Lose it, and the funhouse mirror holds. The most important fight for human freedom in the next twenty years might be that small, and that technical.

* * *

I could be wrong about a lot of this. Who am I, safe and far from any of it, to say my way is right for a woman in Tehran or Riyadh. Maybe some of what I hope spreads is just my own preference. I hold that loosely.

And here is the thing that took me longest to reach. Maybe the point was never to decide whose way of life is right. Look at what we fight over. One country walls its people in and calls it protecting our way of life. Another pushes its values out and calls it spreading freedom. Underneath both, the same human thing is happening. Everyone is clutching a home. Everyone loves their food, their language, their family, the way their grandmother did things. We are identical in the wanting. We only differ in the what.

So the very thing that divides us, everyone guarding their own way, is the clearest proof of what we share. The woman in Tehran and the teenager in Oslo are not enemies across a values gap. They are two people who both love their lives and are afraid of losing them. That is not a difference to defeat. It is a sameness to recognize.

I do not want the world to end up the same. That would be a loss. I want us to stop ranking each other. Your way is not better than mine and mine is not better than yours. The space between us was never a battlefield. It is where we could learn from each other. The thing everyone is protecting, their own way of life, could be the very thing that brings us together, the moment we notice we are all protecting one.

There is only one thing I hold with a closed fist, and it is not a culture at all. It is the girl’s right to choose her own. When the Taliban forbids the girl in Kabul to read, the wrong is not that one culture beat another. The wrong is that she never got to pick. Freedom is not one more flavor competing with the rest. It is the soil all the flavors grow in. Protect the soil, and let a thousand ways of being human bloom, none of them the winner.

* * *

So where does that leave me. Not with a promise. The short road runs through fire, and some of the fuses AI lights will end in a graveyard, not a free country. I will not pretend otherwise.

But maybe, in hard ground, a root is being planted. The girl in Kabul learning to read from a phone. The woman in Tehran hearing, for the first time, that she is not the only one who doubts. A mother somewhere earning her own money for the first time in her life. A little more each day.

And maybe that root gets enough water. Maybe each year puts out one more bloom. One more girl who can read, one more woman with a voice, one more life nobody handed her and nobody can take back. A little more of the world curious about its differences instead of at war over them.

I cannot prove that. I know how naive it can sound on a hard day. But I would rather plant the root and water it than stand over the dirt and swear nothing will grow.

So here is what I will leave you with. Somewhere tonight, a girl who was told she could not learn is holding a phone that says otherwise. What is the one small thing you could do, from wherever you sit, to help one more hand reach one more open door? That is how roots get water. One at a time.

* * *

Sources

Taliban ban on girls’ education past the sixth grade (2021, reaffirmed 2022). PBS NewsHour: https://www.pbs.org/newshour/world/under-taliban-rule-afghan-girls-end-the-sixth-grade-in-tears-their-education-is-over | UNESCO: https://www.unesco.org/en/articles/afghanistan-four-years-22-million-girls-still-banned-school

Mahsa Amini’s death in the custody of Iran’s morality police (2022), and the UN finding that Iran was responsible. Amnesty International: https://www.amnesty.org/en/latest/news/2023/09/what-happened-to-mahsa-zhina-amini/ | UN OHCHR: https://www.ohchr.org/en/press-releases/2023/09/iran-one-year-anniversary-jina-mahsa-aminis-death-custody-heightened

Saudi Arabia lifting the ban on women driving, June 2018. CNN: https://www.cnn.com/2018/06/23/middleeast/saudi-women-driving-ban-lifts-intl | UN News: https://news.un.org/en/story/2018/06/1012962

Censorship and state narratives built into Chinese AI models, and China’s 2023 rule requiring generative AI to uphold core socialist values. Reporters Without Borders: https://rsf.org/en/controlling-information-age-ai-how-state-propaganda-and-censorship-are-baked-chinese-chatbots

Syrian civil war death toll (hundreds of thousands) and displacement (millions). Our World in Data: https://ourworldindata.org/data-insights/the-syrian-civil-war-has-killed-hundreds-of-thousands-displaced-millions-and-caused-poor-health-and-widespread-poverty

Killings, mass arrests, and executions during Iran’s 2022 Woman Life Freedom protests. Amnesty International: https://www.amnesty.org/en/latest/news/2024/09/iran-two-years-after-woman-life-freedom-uprising-impunity-for-crimes-reigns-supreme/ | Human Rights Watch: https://www.hrw.org/news/2025/09/16/iran-impunity-reigns-3-years-after-crackdown-on-protests

For a long time I thought a mind like that was a gift. You see the pattern before other people do. You arrive early. It is a little like watching a movie in slow motion, catching every frame while everyone else sees one smooth blur.

Then try living a life with it.

I have had twelve jobs in fifteen years. For a long time people asked me why I could not settle, and I never had an answer that felt true. I called myself restless. I quietly wondered if something was wrong with me. The real answer took me years to find. I am a builder. I come alive making something new, learning it, leaving my mark on it. And the moment it is built, when all that is left is the daily upkeep of something already figured out, it goes from thrilling, to mundane, to, honestly, torture. I was not flaky. I was a builder being asked to be a maintainer, in a system built for the opposite kind of mind. The same drive that makes me leave once the newness is gone is the same drive that lets me start a thing from nothing. Gift and curse. The same wire.

I am also dyslexic. I was the odd duck in school, in a building designed for a brain that was not mine. And the exact wiring that made reading hard the way they wanted it done is the wiring that reads patterns no one ever taught me to see. Nobody sat me down and told me my mind was not worse, just different, and a little early. I had to find that out myself, the slow way.

Living early is lonelier than it sounds. You learn to wait. You learn to help by asking questions instead of giving answers, because handing someone the ending robs them of getting there themselves. You learn to hold the thing you see quietly, because the one time you said it too soon at the dinner table, the whole table went still, and you felt the door close in your own house.

And here is the thing this whole piece is really about.

No one is ever better than anyone else.

We do not run the same software at different speeds. We run different software entirely. Cognitive profiles vary more than most people will ever realize, and most people have never once stopped to understand their own. The mind that arrives early is not better than the mind that stays fully present. I can sketch a plan for a country I cannot fix, and I cannot do the thing the people I love do without even trying, which is simply be at the beach, in their bodies, in the day, not three layers underneath it. My gift is also my loneliness. Their ease is a gift I do not have. Different wiring. Equal worth. Every time.

Every gift is also a curse. I love hard. I feel what the people around me feel, often before they say it. It lets me show up for people in ways I am proud of. It is also a weight I do not always know how to set down. I carry feelings that were never mine to carry, and some days I cannot tell where other people end and I begin. There is no profile that is only gift. There is no person who is only ahead. Once you really see that, the whole ranking collapses, and something kinder is left standing.

So it comes down to a choice. A few of them.

The choice to understand your own mind instead of judging it against a standard built for someone else’s. And then to extend that same grace outward, to meet people where their brain actually is, and where they actually are in the things that are genuinely in their control.

The choice to lead with kindness, curiosity, and openness, and to make those the heart of how you move, because that is the heart of being human. To set down judgment. To sit with hard truths instead of flinching. To say I love you when someone is scared, even when it is not the expected thing. I told my daughter’s teacher I loved her once, out loud, and I do not regret it. It was simply my truth, so I said it.

The choice to be daring. To question the norms we never actually agreed to. To never, ever stop being yourself, for any person or any room that asks you to shrink.

Because underneath every different mind, under all the wildly different wiring, the human thing is so much simpler than we make it. We are not as different as we fear. We all want the same thing. We all want to be seen.

And now the part I can only say if I am being fully honest with you.

Some days I cannot accept myself for who I am. I know all of this. I just wrote all of this. And there are still mornings I look at the wiring I spent a whole essay calling a gift, and I feel like too much, or not enough, or both at once.

It is a work in progress. There are days I think I have reached some intermediate level of making peace with my own mind, and then, humbly, I fall right back to my knees at beginner. I used to think that meant I was failing at it.

Maybe the thing in front of all of us is that it was never linear. You do not graduate from being human. You do not reach self-acceptance and then keep it on a shelf like a trophy. You climb, you slip back, you climb again. The growth is a spiral, not a ladder, and falling back to beginner is not proof you are bad at it. It is just the part of the path that bends down before it turns up again.

So if today you fell back to beginner too, at accepting yourself, at being kind to the wiring you were given, at believing you are worthy exactly as you are, you are not behind. You are right where the path actually goes. I am there with you, more often than I let on.

I see you. And on the days I forget how, I hope you will see me too.

It rolled across the field on its own. It scooped up the golf balls. It has learned the shape of the range. It knows when the field is full. It knows when to go back in. No driver. No breaks.

There was still one man out there. One person in a little truck, doing the same job the old way. For now.

That job used to take six people. Not six at once. Six to cover the whole week. The place is open seven days, from morning to late at night. So it took a team, taking turns, to keep that one truck running.

It is about three people now. Soon it will be fewer. And the last person standing will not be driving a truck. He will be telling the machines what to do.

This is the part most of the AI talk gets wrong. The job does not just vanish. It shrinks. It changes shape. It goes from doing the work to running the thing that does the work.

But hold onto that man in the truck. We are going to come back to him. Because there is a harder truth hiding inside this story, and most people never get to it.

Let me show you how this really works. In plain words.

The picture most people have

Most people picture an office.

Picture a team of four graphic designers. They make the ads. They make the logos. They do the work by hand.

Now the company buys AI tools. One designer learns to use them well. She does not draw much anymore. She tells the AI what she wants. She picks the good one. She fixes what is wrong. She has taste. She has judgment.

The company looks at the team. One person plus AI is doing what four people used to do. So the team goes from four, to three, to two, to one.

That part is real. But it is only the start of the story.

What AI is actually good at

Here is the thing almost no one explains.

AI is very good at one kind of task. It is good at things that repeat. It is good at things that follow a pattern. It is good at work that does not need much judgment.

Show it the same kind of problem a thousand times, and it learns the pattern. Then it can do that task again and again. It does not get tired. It does not get bored. It does not have a bad day.

So look at any job and ask one question. How much of this job is the same thing over and over?

If the answer is “most of it,” that job is in danger.

It does not have to be better. It has to be cheaper.

Here is where most people are wrong.

They think the machine takes the job because it is smarter than you. It is not. Not yet. It still makes mistakes. It still needs a person to check it.

It takes the job because it is cheap. And because it is good enough.

One person plus the machine can do what six people did. The machine does not have to beat you. It just has to be close enough that the math works. Cheaper, and good enough. That is all it takes.

That should scare you more than “the robot is a genius.” Because the robot does not have to be a genius. It just has to be a deal.

The safe spot keeps moving

People will tell you to get to the “thinking” jobs. The judgment jobs. The ones the machine cannot do.

That is good advice for today. But be careful. The line keeps moving.

Last year the machine could not write the email. This year it can. Last year it could not make the picture. This year it can. The work that needs a human keeps getting smaller.

The safe spot is not a rock. It is a sandbar, and the tide is coming in.

So the real skill is not “get to the safe side and stop.” It is “keep moving as the line moves.” Stay the person who runs the machine, even as the machine gets better.

The part most people miss

Most people think the first wave is only office jobs. The designer. The clerk. The customer service rep clicking the same five screens. The accounts clerk paying the same kind of bill ninety times a day. AI can do all of that right now.

But that is only half of it.

The same thing is already hitting hands-on jobs. Outside jobs. Blue collar jobs.

The man in the truck at Top Golf. The robot does his route now.

At Costco I saw a lawn mower that drives itself. You walk it around your yard one time. It learns the shape of your lawn. After that it cuts the grass on its own. No person pushing it. It is like a Tesla that drives itself, but for your yard.

The lawn crew. The pool guy. The same kind of route. The same kind of pattern. Done by a machine that learns it once and never forgets.

But the two waves move at different speeds

Here is something almost no one says.

The office wave and the outside wave are not the same speed.

An AI tool is just code. You can copy it to a million computers overnight. That is why office jobs are changing so fast.

A robot is a real thing. Metal and wheels and batteries. To put a robot in every yard, someone has to build it, ship it, fix it when it breaks, and deal with rain and hills and dogs. That takes money and time.

So the outside wave is real, but it is slower. It comes lumpy. A mower here. A ball machine there.

Code moves fast. Steel moves slow. Both are moving.

Why it feels so loud right now

This is why the whole world is suddenly yelling about AI.

The first wave is hitting the loud people first. Designers. Writers. Analysts. The people who post and write and make noise. So you hear about it everywhere.

The quiet wave, the one coming for the lawn crew and the pool guy and the man in the truck, is just as real. But those folks are not writing the think pieces. So you do not hear about it yet.

I am telling you about the quiet half because I saw it with my own two eyes. At Costco. At Top Golf. Before the news did.

Now follow the money

Here is the question almost no one asks.

When six jobs become one, the work still gets done. The grass still gets cut. The balls still get picked up.

So where did the money go?

It did not vanish. It moved up. It went to whoever owns the machine. The person who sold their time got less. The person who owns the tool got more.

That is the whole story in one line. The money is moving from people who sell their time to people who own the machines.

Watch it happen in your own neighborhood. The lawn company you hire gets smaller. Or people just buy the machine and skip the company. Or someone builds a new company, buys a hundred machines, and sends them out across town. One person, watching a screen, running all of them. And the money that used to feed a crew now feeds one owner.

What “repricing” means

There is a word for this, and most people have never heard it. Labor repricing.

It sounds fancy. It is simple.

It means the work is worth less money than it used to be.

When a machine can do most of a job, the human part is worth less. So the pay goes down. The company posts fewer of those jobs. The raise gets smaller. That yearly bonus you were counting on, the one that comes every December, do not count on it.

Nobody knows the exact day

One more thing, because I want to be straight with you.

I cannot tell you the date. Anyone who gives you a date is guessing. Some companies tried AI, it did not work yet, and they hired people back. The road is bumpy.

So do not trust the timeline. Trust the direction.

The mower is coming. The ball machine is already here. I will not tell you the month. I will tell you the way the river is flowing.

So here is the honest part

Now back to the man in the truck.

I told you to learn to run the machine. To be the one who stays. And you should try.

But I have to be honest with you, because you trusted me this far.

There were six of them. There is room for one. Five of those six cannot all become the boss of the robot. The whole point of the robot is that it does not need them.

So this is not only a “work harder” story. You can do everything right and still get swept up, because the chairs themselves are going away.

That is why this is bigger than you. It is bigger than your plan. It is a change in who gets the money in this whole country. And a change that big is not something any one person can hustle their way out of.

So do both. Make your plan. Learn the machine. Be glad for the job you have today. Truly.

And start paying attention to the bigger thing. Because this one we are going to have to face together. Not one truck at a time.

I am not worried about the machine on the golf range. I am thinking about the man still out there in the truck. And about the five who used to be out there with him.

Which one are you? And what are we going to do about the ones who are already gone?

It is a fair question. It is also a little like standing in a house that is filling up with water and asking if the carpet is going to get wet. Yes. The carpet. Also everything else.

Here is what I mean. We keep talking about AI like it is a faster tool for getting work done. A really smart calculator. A really fast intern. And it is those things. But that is the smallest part of what is happening, and it is the only part most of us are looking at, because it is the part with a paycheck attached.

The bigger thing is quieter. AI is not just changing how we work. It is changing how we think. How we figure out what is true. How we understand ourselves. How we talk to each other. None of that shows up on a pay stub, so almost nobody is watching it. But that is where the real change lives.

Let me make it simple. Think of AI like a machine you can use two completely different ways. You can use it like a gym, where it makes you stronger. Or you can use it like a couch, where it does all the work and you slowly get weaker. Same machine. The difference is not the machine. The difference is you.

And here is the uncomfortable part. The couch is easier.

We actually have proof of this now, and it is a little jarring. In one study this year, researchers had AI give people wrong answers on purpose. People went along with the wrong answer about eighty percent of the time. Only one in five stopped and said, wait, that does not sound right. Eighty percent just nodded and kept going. The researchers gave it a name. They called it cognitive surrender. (I know. Cheerful.)

Another study found something even bigger. When people use AI passively, just taking whatever it hands them, it weakens the brain over time. When people use it actively, arguing with it, checking it, building on it, it makes them sharper. Same tool. Two opposite outcomes. And the thing that decides which one you get is whether you stay in charge of your own thinking, or hand it over. That right there is the whole ballgame.

Now here is what almost nobody is connecting. This is not just about work, or just about thinking. It is the same fork showing up everywhere at once.

How you figure out what is true. Stay in charge, or surrender. How you understand who you are. Use it as a mirror, or let it tell you. How you connect to other people. Let it help you say what you mean, or let it speak for you. And what your kids become, which is the one that should stop us cold, because kids are using it the most, depending on it the most, and their thinking is the most up for grabs.

Every one of those is the same question wearing a different outfit. Do you stay the author of your own mind, or do you hand the pen to the machine.

And I have to be honest with you, because pretending otherwise would not help. Handing it over is the easy path. It is the couch. It is the eighty percent. Letting the machine think for you feels like relief, right up until you realize you cannot do the thing yourself anymore. Staying sharp takes a little effort every single time. It is a choice you have to keep making, on purpose, against the easy version.

So no, I do not think the big story is whether AI takes your job.

The big story is this. For all of human history, our tools helped our bodies. The shovel for our back. The wheel for our legs. The telescope for our eyes. This is the first tool ever that does the one thing we always had to do for ourselves. It thinks. And thinking, the real kind, the deciding and the questioning and the figuring out, is not a side feature of being human. It is most of the job.

So the question underneath all the noise is not really about robots at all. It is a very old, very human one, just asked in a new way. Are you going to stay the author of your own mind.

You do not need to be a genius to answer it. You do not need to understand how any of this works under the hood. You just need to do one small thing the next time you use it. Do not just take the answer. Poke it. Ask it why. Decide for yourself if it is right.

That is the whole practice. That is the gym instead of the couch.

Stay in there. Keep your pen.