Hi everyone. It’s good to be back.

A lot has happened since I last emailed you. New wars. New models. New headlines declaring salvation or collapse depending on the day of the week. I published a book. Thousands of new people subscribed to this newsletter. Somewhere in the middle of all that, the conversation around AI became much louder and somehow less useful.

Before I go further, I want to first reaffirm the purpose of this newsletter.

When I started writing online, I strictly did it for the purpose of testing my own ideas in public. The internet is useful that way. Once an idea is written down under your own name, you have to live with it, which forces a certain conviction and clarity.

But as the audience grew, something subtle changed. I found myself writing to educate people rather than to discover what I actually believed. Gradually, I saw the exercise more as a public service — I was defending hope and optimism by documenting how AI (and technology, broadly) was expanding human potential.

So over the last year (and especially over the last 3 month hiatus) I started asking many of you what resonated most. The answer was surprisingly consistent: we like when you tell us what you really think, even when it might be hard to hear.

That was galvanizing for me, because it turns out that’s what I truly love writing about.

It was the lens through which I wrote my book, The Next Renaissance (which you can now listen to on audio as well as print).

I wanted it to be hopeful without being naive; ambitious without becoming science fiction; understandable without flattening the complexity.

What I didn’t know, while writing it, was who it was actually for.

That uncertainty created constant tension. Every chapter raised the same question: who is this helping?

Then my daughter Frankie was born — and I immediately knew.

I realized I hadn’t written it for technologists or executives or policymakers, at least not primarily. I had written it for parents. Or more broadly: for anyone responsible for another human being, trying to make sense of a world that is changing faster than they can explain it.

That realization reframed almost all of my work.

On stage, in boardrooms, at workshops, town halls, union meetings; everywhere I go, I increasingly meet two versions of the same person.

One wants to understand how AI will affect their business — how they can leverage or invest in AI to make more money.

The other is worried that it will come at the expense of the world their children are inheriting.

It’s not difficult to understand the anxiety. Despite a long, proud history of technology improving the human experience, our recent relationship with it isn’t great.

Cell phones captured the soul and spirit of children and stole their innocence. We gave everyone access to unmetered information before we taught them how to metabolize it. Every war, catastrophe, corruption scandal, and existential threat now arrives instantly, continuously, and algorithmically amplified. Social media discovered that outrage spreads faster than truth and built trillion-dollar businesses around the observation.

At the same time, many people watched the foundations of adult life become economically absurd. Housing. Education. Healthcare. The three pillars of a developed society became much more expensive, while seemingly everything else became frictionless and cheap.

And now we are entering the era of unmetered intelligence.

In 2021, my friend Boris Power and I wrote a paper called Unmetered Intelligence. The thesis was simple: intelligence itself was beginning to behave like a utility. Something abundant. Cheap. Ambient. Like electricity or bandwidth.

That prediction now feels less theoretical by the month.

The cost of advanced cognition is collapsing. Models that were prohibitively expensive two years ago are now effectively free.

This matters for all sorts of reasons; principally, we should all care about the decline in cost of anything, because civilizations change when essential resources become abundant.

But abundance is morally neutral.

Access to the internet does not make everyone thoughtful. Literacy does not guarantee wisdom. Access to gyms doesn’t guarantee fitness.

What we choose to do with it will define our outcomes, individually and collectively.

The dystopian version is already familiar. Extreme outcomes include supercharged villains, economic collapse, extreme concentration of power.

More insidious pictures describe passive people, synthetic relationships, outsourced meaning, algorithmic sedation; a civilization that automates so much of life that it eventually forgets why life was valuable in the first place.

On the other hand, the optimistic version is usually described without much clarity either: AI as an omnipotent god; paradise found; struggle solved; heaven on earth.

All of these outcomes present this technology (and seemingly future technology) as something that happens to us, instead of for us, or with us.

Neither future is inevitable. Technology does not absolve us of responsibility; it concentrates it.

Going forward, this newsletter—now renamed Unmetered Intelligence—will explore the promises and dangers of abundant intelligence.

Some essays will be practical. Others will be philosophical. Some will probably be unpopular. All will attempt to frame constructive responsibility as a reasonable alternative to ambient dread, even when one is much less appealing (or even believable) than the other.

I’ll try to publish twice a month. Occasionally more when something important happens, or less when the weather in Santa Barbara is particularly good.

It’s good to be back. See you again soon.

I wrote a book.

The Next RenAIssance: AI and the Expansion of Human Potential is officially out today!

After spending the last three years in more than 250 boardrooms and on stage in front of roughly 300,000 people, I wrote this book as an answer to the question I’m asked most often, in one form or another:

What happens next?

I’m humbled, thrilled, and overwhelmed by this moment. If you’re one of the 25,000(!) people who pre-ordered it, I offer my most sincere thank you. If you have not yet, I hope you’ll consider ordering a copy today.

On behalf of an industry that so often struggles to explain itself, the book is my best attempt to help the average reader (my mother) make sense of this moment — and what may follow — by placing artificial intelligence in its proper historical context, offering a clear-eyed view of its costs and benefits, and proposing actions we can take to make the most of it.

I hope you, and perhaps someone in your life a bit less tech-savvy, can learn something new and look toward the future with renewed optimism.

A Promise to the Next Generation

While this book is a culmination of my career in AI, the motivation behind it is deeply personal.

The journey began around the same time I learned I was going to be a father. That realization shifted my perspective entirely. The “future” was no longer an abstract concept or a timeline on a slide deck; it was a place where my child was going to live. When we welcomed my daughter into the world this past October, I knew immediately that I wanted to help create a world worthy of her.

I wrote this book because I believe that being optimistic, aiming for massive societal successes, and thinking deeply about how we can improve the world for everyone is the most worthwhile endeavor a parent can undertake. I want my daughter to grow up in a reality where technology expands human potential and happiness, rather than diminishing it. To do that, we can’t afford to give room to cynicism.

How AI Helped Me Find the Soul of the Book

Writing a meaningful book is a humbling, satisfying, and incredibly difficult endeavor. There are already things I’d revise. But as I look at the finished product, I’m not sure I could have done it without AI.

To be clear: AI did not write this book.

However, AI made the book possible. It acted as the ultimate research assistant and structural editor. It helped me get my ideas straight, organized my chaotic thoughts into coherent frameworks, and managed the data so I could focus on the message.

By offloading the organizational and structural heavy lifting to AI, I was granted a rare luxury: the time and mental space to do the human work.

It cleared the path for me to focus on storytelling, on weaving in human dreams, and on connecting with the humans who would read it. AI gave me the freedom to put soul into the work.

The Path Forward

I’m proud that these pages reflect what I’ve learned from a career in AI, years studying the history and possibilities, time with my family, and conversations with thousands of people — world leaders, business leaders, and community builders.

The story of technology is, ultimately, a story of progress. To make sure that you progress with it, I’ll leave you with my four principles for living a better life in the age of AI:

1. Go Outside: The physical world matters more than anything else. Be in it.

2. Be Human: As intelligence becomes abundant, human qualities become more valuable.

3. Learn How to Learn: For most people, the most important skill in a changing world is the ability to acquire new ones.

4. Lead with Optimism: Not because a better future is guaranteed, but because optimism is a critical ingredient in building one.

Thanks for being here; I hope you’ll consider reading The Next RenAIssance: AI and the Expansion of Human Potential.

Prophesizing any single event more than a year out is hard; accurately predicting further than that is a fool’s errand. As a result, bold predictions (especially those made loudly) are mostly just personality tests.

Even so, people often ask me what I think the future holds, and I’m inclined to provide an answer.

So here you go: I believe that unmetered intelligence — cognitive capacity at near-zero cost — will deliver a few trends in 2026:

1. Roger Bannister Moments — the exposure of false boundaries

2. Three Mile Island Moments — the reckoning of major downside events

3. A renewed hunger for, and investment in, the physical world

1. We Will See Multiple "Roger Bannister Moments”

For centuries, physiologists believed that running a four-minute mile was physically impossible. They believed the human heart would explode, muscles would tear from the bones, and athletes would collapse as they pushed past their limits. Runners mostly believed this as well, adding a psychological boundary to the perceived physical one.

Then, on May 6, 1954, in Oxford, England, Roger Bannister ran the mile in 3:59.4.

Roger Bannister will forever be remembered as the man who revealed a false boundary. Even more impressive is the fact that his record stood for only 46 days. In less than two months, Australian John Landy broke Bannister’s record by a second and a half. Within three years, 16 men had run a four-minute mile. When Bannister proved that the physiological and psychological limits were false boundaries, he opened the floodgates.

Bannister’s accomplishment went beyond fitness, it exposed a deeper truth about human performance: belief changes what the body is willing to attempt.

Once a limit moves from impossible to done, effort reorganizes around the new ceiling. This pattern appears well beyond running. The placebo effect shows that expectation alone can measurably alter pain tolerance, endurance, and even physiological markers. The Pygmalion effect demonstrates that people perform better when more is expected of them. And decades of research on self-efficacy, led by Albert Bandura, show that individuals who believe success is attainable persist longer, tolerate discomfort better, and recover faster from failure. When belief updates, behavior follows. Progress, more often than we admit, is unlocked by the realization that the old limit was never real.

Recently, we’ve seen several false boundaries collapse. For decades, a diagnosis of cystic fibrosis (CF) meant a sharply shortened life; many patients didn’t survive past their teens. That understanding unraveled with the approval of a triple-drug therapy in 2019 — effective for roughly 90% of people with CF — and with additional therapies moving through laboratories in 2025. As a result, newborn life expectancy for CF patients has risen to levels approaching that of the general population. Human biology didn’t have to change — our ability to navigate it did.

AI and unmetered intelligence will produce similar “Bannister Moments” across industries by revealing many presumed limits as resource constraints, rather than physical constraints. These false boundaries tend to fall into three categories: soft boundaries, imposed by cost or time; social boundaries, reinforced by the collective belief of what “can’t be done”; and interface boundaries, where answers exist but are buried in complexity beyond human cognition.

AI is an aggressive solvent for all three — and we already have the evidence. Drug discovery once depended on years of trial, error, and luck. In 2025, researchers at MIT used AI to invert the process, screening 36 million compounds to identify candidates effective against drug-resistant bacteria. At NJIT, researchers applied AI to discover new porous battery materials, bypassing the physical bottleneck of testing millions of material combinations by hand. And at UNSW, machine learning reduced more than 8,000 potential green-ammonia experiments to just 28 high-probability targets — turning years of failure into weeks of focused validation.

The same dynamic plays out at the individual level; people carry their own false boundaries — beliefs about intelligence, creativity, leadership, or resilience — that feel permanent only because they’ve never been bested. But once an individual sees proof that things might not be as they seem, behavior changes: I can do this. Effort increases. Risk tolerance expands. Persistence follows. Psychologists describe this as self-efficacy; most people experience it as growth. It is the same mechanism that drives societal progress: beliefs update in response to evidence, and capacity expands as a result.

In 2026, we will start to recognize that some of our most stubborn limits — personal and collective — were never laws of physics, only limits of human bandwidth. As discovery shifts from lucky breakthroughs to reliable, AI-driven search, progress in longevity, clean energy, heavy industry, and personal agency will accelerate — and what once felt miraculous may start to feel… routine.

2. The Industry Faces a "Three Mile Island Moment”

Failures have already occurred during AI’s rapid expansion: hallucinated legal citations, automated systems reinforcing bias, deepfakes used for fraud, AI-enabled scams at scale, and opaque models embedded in consequential workflows without sufficient oversight. AI’s public perception has largely survived these incidents because their costs so far have been diffuse, reversible, or absorbed quietly.

The risk ahead is not the first failure. It is the first highly visible, emotionally resonant failure — one that is confusing, frightening, and public. When that happens, even if the objective harm is limited, the public response will register as a crisis of trust.

History shows that losses of trust often appear as consequences of failures in governance, containment, alignment, or deliberate misuse — not sudden revelations about technology itself. That asymmetry is the lesson of the Three Mile Island accident of 1979, where a partial nuclear meltdown released only small amounts of radiation, and caused no immediate deaths, but created an outsized response.

On a risk-adjusted basis, it was a relatively low-cost event, but it was public, opaque, and terrifying. Operators appeared overwhelmed. Systems were poorly instrumented. Communication failed. The result was not a proportional response to risk, but a collapse of public permission. Nuclear progress stalled for decades because institutions failed to contextualize failure and manage fear — not because the technology was uniquely dangerous.

When we consider AI and the potential for failure, we should be precise about what that next incident is likely to look like. It will not resemble science fiction. More plausibly, it will involve:

A system behaving exactly as designed, yet misaligned with human intent:
Consider a hypothetical national health insurer that deploys an AI system to standardize care approvals and reduce unnecessary procedures. The model performs well — costs fall, variance shrinks. Then journalists discover that hundreds of patients across multiple states had chemotherapy, cardiac procedures, or neonatal care delayed or denied because the model optimized for population-level cost efficiency rather than individual medical urgency. No malfunction. No rogue model. But the human intent — to provide care first and manage costs second — was quietly reversed at scale.

A governance breakdown where responsibility is fragmented or unclear:
Imagine a government agency rolls out an AI system to help allocate disaster relief after a major hurricane. Aid is delayed for weeks in several regions. Local officials blame the federal algorithm. Federal agencies blame contractors. Contractors point to incomplete data. Congressional hearings follow — and no one can explain, in plain language, why help didn’t arrive.

A containment failure that allows a small error to propagate:
If a cloud provider’s AI-powered traffic optimization system misclassifies a routine software update as a security threat, the resulting problems could ripple far into daily life. In response, it could automatically reroute and throttle traffic across critical services — payments, logistics, emergency dispatch — pushing millions of users across multiple countries “offline” for hours.

Deliberate misuse by a bad actor:
During an election cycle, if a coordinated group uses generative AI to flood social media and messaging apps with hyper-local, highly credible deepfake videos of election officials announcing polling place changes, the fallout could be massive. The misinformation could be corrected within a day, but voter turnout might drop measurably, lawsuits could follow, and confidence in the election results may not be restored in timelines that matter.

Critically, in each of these cases, the perceived risk may vastly exceed the actual harm.

The downstream cost of the Three Mile Island accident is often overlooked. Energy demand did not disappear. Instead, it was met largely by coal and natural gas. Over the following decades, the world burned hundreds of billions of tons of coal, contributing to air pollution responsible for millions of premature deaths and to greenhouse gas emissions that now define the climate crisis. In hindsight, the alternative to nuclear power proved far more damaging — both to human health and to the planet — than the technology that triggered public fear in the first place.

Most experts (even the most fear-mongering) agree that we are fairly far away from an AI-driven catastrophic failure. But we are certainly not far away from a failure big enough to frighten the public, one complicated enough to be difficult to explain, triggering a reaction wildly disproportionate to the damage done. On a risk-adjusted basis, AI systems may still outperform human-led alternatives, but that may not matter.

Every foundational technology pays a tax during real-world deployment. The question is not how to eliminate failure — that is impossible — but how to limit its cost, explain it clearly, and prevent a low-damage, high-visibility incident from freezing progress altogether.

3. The Human Renaissance

In 2026 people will spend, per capita, more time outside and in physical communities.

One reason for this shift is growing frustration with the proliferation of “AI slop” across the internet. The other is our realization that screens are making us all feel terrible.

The developed world is getting very close to finally, collectively, pointing to the screen (and our addiction to it) as the principal culprit of our recent decline in happiness. The fight against screen time, especially among youth, will lead to a renewed interest in the physical world.

As Jonathan Haidt famously detailed in The Anxious Generation, the attention economy’s “rewiring” of our minds has triggered a global mental health collapse. We are digitally saturated but socially starving. Reports from the U.S. Surgeon General and the World Happiness Report link smartphone ubiquity directly to skyrocketing anxiety and loneliness.

We know how to solve this, because we know what makes us happy. The Harvard Study of Adult Development proves that face-to-face relationships are the primary predictor of health, while evolutionary psychology confirms our nervous systems are designed for complex natural terrain, not scrolling feeds.

In 2026, we will see a rise in real life, physical experiences. We’re already seeing the start of this trend. Pollstar’s 2025 Analysis shows that while live music ticket prices stabilized, crowd density exploded — Top 100 Worldwide Touring Artists sold an average of 19,104 tickets (up 11.9%), an all-time record. The Broadway season defied all expectations with a historic $1.89 billion gross, and MLB attendance grew for the third straight year, drawing 71 million fans.

People are turning to the outdoors at record numbers; 181.1 million Americans reported recreating outdoors — 58.6% of the U.S. population — and the National Parks reported a record-breaking 331.9 million visitors in 2024 (our latest full year with visitor stats).

We have a human drive for community and the outdoors, and in 2026 we’ll return to it. My call to you is to support this movement by telling your local politicians that we need new parks, safer bike lanes, and true “third places” where people of all ages can gather safely and freely — ensuring that the antidote to digital isolation remains accessible, public, and human.

What Can You Do

With these trends in mind, I want to offer three strategies for moving forward in 2026, working with AI, and not losing the qualities that make life worth living.

1. Tell Stories About a Better Future: Pilots and racecar drivers use the term target fixation to describe the phenomenon that says the more you look at something, the more likely you are to move towards (or into) it. Acknowledge the danger, and look towards the safe, abundant paths ahead. We create the futures we imagine. Aim at the good ones.

2. Agency is Your Superpower: Personal responsibility is quickly becoming a major determinant of success, as more and more people have access to more supercharged tools and technologies. In an era of unmetered intelligence, the advantage belongs to those who choose to care, take responsibility for their growth, and act.

3. Be Human: You can’t beat a machine on its terms. As they get better at analysis, prediction, and execution, invest in stubbornly human attributes: courage, compassion, curiosity, humor, wisdom, and hope. These are our new advantages. And life is better that way, anyways.

One of the most corrosive trends in our AI discourse isn’t technical — it’s linguistic.

Phrases like “to protect our humanity, humans should always be in the loop” get thousands of likes because they sound wise. They’re not. Humans do not need to manually drive cars to preserve their humanity. These phrases offer comfort rather than insight.

This kind of slop (what I’m calling linguistic imprecision) is fueled by algorithms that reward banality, the genuine complexity of AI, and something older and subtler: the slow degradation of meaning itself.

It’s causing us to misunderstand (and vilify) essential parts of AI.

Denotation vs Connotation

When you do something the wrong way enough, it can start to feel like the right way. It’s as true in seasoning food — I grew up in a household that didn’t cook with salt or butter, which I now see is insane — as it is in language. In fact, there are terms to explain this lexical phenomenon: “denotation” describes what a word means, and “connotation” describes what people want it to mean.

Connotation doesn’t just change how we use words it changes how we understand the world through them.

Terms like “ignorance,” “activist,” “liberal,” and “conservative” are rarely used to denote their literal definition. They’re used to connote a figurative definition, in many cases as a pejorative (a word expressing contempt or disapproval).

“Bias” is another such victim of connotation. It has become a nasty word, used to describe behavior like unfairness, meanness, or even bigotry. As early as the 16th century, people used it to describe “undue prejudice” in the legal system. By the early 20th century “bias” was the barb cast at publications and reporters that readers felt (fairly or unfairly) didn’t meet the new standard of objective journalism. In the latter half of the 20th century, the rise of social sciences spread the term into everyday social life, recasting it as a quasi-diagnostic flaw.

But bias is, by definition, “a tendency, inclination, or prejudice toward or against something or someone.” It is, denotationally, a neutral term.

Today, when we hear someone use “bias,” especially if they accuse someone else of it, we can safely assume acrimony: the coach who doesn’t put your child in the game enough, the teacher who grades an essay harshly, the banker who doesn’t approve your loan — they’re all biased.

We don’t hear bias given credit when parents take care of their children (“good parenting” rather than a “bias for care”), when people are nice (“kind souls” vs “biased for warmth”), or when others pick up litter in public places (“acting like a Boy Scout” instead of “biased for civic duty”). Doing good is driven by a bias to do good.

Bias is essential. For humans and for AI.

But the distance between its connotation and denotation is having a major effect on one of the most important issues in AI, and leading many people to believe bias is something that should be eliminated.

Bias as the Basis of Agency

Biases are not defects in human cognition; they are its foundation.

Humans are able to operate in the world because biases turn all our lived experiences and genetic predispositions into a series of impulses (both good and bad). They are the mental frameworks that allow us to quickly and efficiently turn inputs, wants, and needs into thoughts and decisions. They allow us to move through the world with agency, rather than pinballing through it like passive zombies overwhelmed by stimuli.

These cognitive shortcuts give us, in a moment, access to a lifetime’s worth of insights, while protecting us from memories we don’t actually want to consider when we contemplate what flavor ice cream to get.

In AI, “bias” was adopted early as a technical term, along with “weights,” to describe how and what the machine considers and prioritizes when making a decision. My first boss, Lukas Biewald, is actually the CEO of a company aptly called Weights and Biases (recently acquired by CoreWeave), which helps developers build and manage their AI models.

In humans and machines, biases also allow us to work toward the goals we want (what Ziva Kunda calls motivated reasoning), and create strategies for taking action with incomplete information and uncertain futures, “A biased mind can handle uncertainty more efficiently and robustly than an unbiased mind.”

The complication for humans, though, is that we can’t fully see or explain our biases. Every single thing we have ever experienced informs them, and it is all buried beneath trillions of neurons in our brain and nervous system. There is not enough therapy in the world for any individual to fully appreciate or unravel the rat’s nest of how they formed.

This is why seemingly rational people can do irrational things, and why we often lack answers to the questions “Why did I do that?”, “Why do I feel this way?”, and “What was I thinking?”

Moreover, even if we could make sense of everything we believed, we would never admit most of it. A cop won’t tell you they’ve pulled you over because they’re going through a messy divorce, and a waiter won’t tell you they recommend the meatballs because it makes the restaurant smell like their childhood home.

We move through the world with an unspoken understanding that our deepest motivations are often unknowable (though Freud argued otherwise) and many of the known ones (both good and bad) are far too intimate to share.

This leads to one of the stranger elements of the human experience: the fundamental inexplicability of everyone’s decisions (including our own).

Solving the Explainability Problem

When people ask me, “How do we remove bias from AI?” they usually mean one of two things:

• How do we stop machines from amplifying injustice?

• Or, more often, How do we make machines agree with my values?

What they should be asking is, “How do we build an explicable machine?”

We all want AI models that:

1. Have a strong “tendency, inclination, or prejudice toward” positive human outcomes. This is the Alignment Problem.

2. Share all of the information they considered (and how they weighted that information) when arriving at a decision. That’s the Explainability Problem.

Quickly, on alignment: alignment is the work that ensures AI models are focused on human interests/values, and can appreciate how their actions broadly affect those interests. AI companies spend billions on alignment, pushing machines until they act incorrectly, learning why they did, and using these insights to further align models with our desired biases and outcomes. (Last month Anthropic explained the clever method they used to learn how a model was reward hacking).

One of the major challenges in alignment is human: the same minds (ours) that argue and fight in the physical world (and are guilty of ills like racism, sexism, and inequity) will need to do the hard work of agreeing on what positive biases we want. With different value systems spread across the globe and societies, it will be difficult (if not impossible) to agree which ones should be encoded in AI; I call this the North Star Problem.

More likely than not, there will never be a single alignment solution. Instead, we will build standards model developers must adhere to, and governing bodies to hold them accountable when they manufacture misaligned models.

That brings us back to explainability.

Explainability is the process of unpacking the models’ bias in decision-making and reasoning by understanding what data they used, what tradeoffs they made, and where uncertainty entered the process. This is much easier to do with machines than humans because we can see the data stored in a machine (and machines don’t have ego or shame).

A common way to do this is to ask a model to “show its work” by breaking down a decision into many small steps and explaining the reasoning behind each one.

Explainability matters for three reasons:

1. It reveals the biases already in the system, including those we did not intend.

2. It allows systems to be re-engineered toward better outcomes.

3. It builds trust based on understanding — not blind faith.

When machines explain themselves, they give us something human systems rarely do: legibility.

This advantage that can allow us to make opaque, human-centered decision-making architectures transparent; imagine admissions decisions, loan approvals, or court judgments that don’t end in “because I said so.” Systems that don’t just aim to be fair, but feel (and are) fair because their reasoning can be examined, challenged, and improved.

The work is not perfect — some AI biases will remain difficult to trace, just as they are in human minds. But success doesn’t require total understanding, only that artificial reasoning be easier to see and understand than human reasoning.

If we demand explicability, we can use AI bias to confront human flaws rather than amplify them — and build systems that are more accountable, more just, and ultimately more humane than the ones they replace.

Visit zackkass.com to learn more about Zack and get in touch.

If you could automate everything, where would you stop?

I regularly ask this question to audiences, friends, and business leaders.

Often, the first answer is a long pause, then a (knowing and hoping-you-don’t-imagine) laugh.

Once you get past the crude answers (which are better with friends than parents), the exercise is illuminating.

It’s easy to say what you want automated: that you hate email, that filling out the same form at the doctor’s office over and over again is maddening, that you want the airline to know (after so many flights) that you prefer an aisle seat.

But when people start talking about things they’d preserve — walking their kids to and from school, baking bread, practicing the violin — those answers reveal deep truths about each of us. They show what people care about — the challenges, annoyances, struggles, and experiences that give meaning rather than distracting from it, the ephemera of our lives that actually make up the core of it.

After giving their answers, I find that often, business leaders and parents turn the question back on me, “Well, what should I automate?”

The actual answer will differ for every person and business. But to help make sense of the question, I ask them to consider automation through this framework:

1. Does this enhance human agency?

2. Does it deepen trust and connection?

3. Does it sharpen human judgment?

4. Does it expand collective opportunity?

5. Can its outcomes be audited or reversed?

By answering these questions, we can begin to draw the line between those things we are comfortable automating, and those we want to remain in human control. I call that line the automation boundary.

Doing so successfully, and creating automation that amplifies our lives rather than hollowing them out, requires determining what friction in our life is vicious, what is virtuous, and what solutions should be left to humans.

The Miracle of Automation

History shows us that over centuries automation has proven to be one of the most consistent ways we have found to improve human life — even if it often comes in tandem with economic upheaval.

Before Gutenberg created his printing press, the number of manuscripts in Europe, hand copied by scribes, numbered in the thousands. A scribe could copy a few pages a day. The printing press could copy 3,600 pages a day. By 1500, about five decades after the printing press, there were more than 9 million manuscripts across the continent.

Textile mills reduced the price of cotton cloth by nearly a third between 1770 and 1801, and another 50% by 1815, making clothing affordable for the masses (and allowing regular people to take part in fashion, a pleasure long confined to the elites).

In 1900, families spent about 40% of their budget on food. By 2020, through mechanized farming, that number fell to less than 10%.

These strides often came with economic upheaval (mechanized farming shrunk the work necessary to feed the American population from 37.9% of the workforce in 1900 to 1.1% of the workforce in 2017), and this upheaval often brings waves of resistance.

Filippo de Strata, a 16th century monk and scribe complained to a chief magistrate about the arrival of the printing press saying, “They shamelessly print, at a negligible price, material which may, alas, inflame impressionable youths, while a true writer dies of hunger.” The Luddites smashed textile mills in an attempt to preserve their jobs and slow the pace of change.

In October 2024, 50,000 dockworkers walked off the job at ports along the American East Coast calling for a ban on automation, even though it would improve the safety of their jobs (which have a fatality rate nearly 5x the average American job).

Often, this pushback focuses on the economic impact of automation, rather than on the benefits it brings. But throughout history, access, affordability, and safety have won out. The cost of living depreciates, the number of things we can enjoy increases, and fewer people are injured and maimed in the process.

Automation has long improved life.

But today, I believe we are reaching its limits (or at least finding the edge).

The False Promise of Total Automation

The temptation of automation is to remove any inconvenience, struggle, or annoyance. AI and emerging technologies are making this future seem possible; it is not hard to imagine a Seussian (or Rube Goldbergian) future where automation makes life easy. Where machines cater to our every need and existence is contentment and relaxation. But as any teacher or coach will tell you (often during conditioning) there is value in hard work.

Some challenges beat us down. Some help us build ourselves up. I differentiate the two as virtuous friction and vicious friction.

Virtuous friction is the type of challenge that helps us grow, helps us connect with others, and provides defining human experiences. These challenges cultivate meaning and mastery.

Climbing a mountain is not the same as taking a gondola to the top. Teaching your kids the birds and the bees is uncomfortable, but it builds trust and a sense of safety. It’s easier to DoorDash a pound of sugar, but Mrs. Ogden will be so glad you knocked on her door and asked to borrow a cup.

In Self-Efficacy: The Exercise of Control, researchers showed that working hard to overcome struggle teaches you that hard work improves your capability, and gives you a means to improve yourself. Students who overcome struggles reading are more likely to see themselves as capable readers.

Learning an instrument, weightlifting in a gym, teaching yourself to freestyle rap, even when it comes with the potential for embarrassment (not a personal anecdote, don’t worry) add joy and color to our life. Research shows that each of us, intrinsically, seeks out challenges because we find them interesting and satisfying.

In some cases, human inefficiencies are the value-add, rather than the chokepoint to be automated away. It is the reason why so many of us have such a distaste for self-checkout. We want the human connection, the conversation, being recognized and valued, even if it means waiting in line behind four carts instead of breezing through a self-checkout with only our two onions.

And crucially, some struggles connect us more deeply to what it means to be human. Science is beginning to show that enduring pregnancy and childbirth adds value to both the mother and child’s life. Natural labor, birth, and early breastfeeding are associated with surges of oxytocin, which is linked to maternal caregiving, stress reduction, and bonding behaviors. Some newer physiological reviews are even exploring if labor and birth causes long-term adaptations to the oxytocin system, further influencing the mother and child interaction.

Vicious friction, on the other hand, is the type of challenge that only diminishes our lives. These hold us back, prevent us from growing, and isolate us from our peers and opportunities.

There is little to be gained from being unable to reach a doctor, not having access to warm and safe environments to learn and grow, or access to utilities that make hygiene the standard.

An overwhelming onslaught of administration that prevents people from accessing (let alone applying for) public benefits does not lead to growth; it keeps families from realizing the assistance meant to help them.

Financial disparity that creates inequitable access to education (often through differences in property taxes) does not teach children to work through educational challenges; it creates artificial burdens that limit opportunities.

Time poverty, often mediated by lower socio-economic status that requires heads of house to spend more hours working to support a family than they spend with them, does not build stronger social bonds; it only compounds existing economic disadvantages.

Frustration that comes from sounding out a difficult word in a second-grade classroom is beneficial. The difficulty of not being able to go to school because you have to walk to gather water is anything but (around the world, women and girls spend 200 million hours a day collecting water, cutting into time for education).

The automation boundary should protect virtuous friction, and eliminate vicious friction.

The Danger of Dehumanization

While we consider what challenges we want to preserve, we should also consider the solutions we don’t want to replace (at least not with a machine). In 1984, Pope John Paul II wrote, “Human suffering evokes compassion.” While eliminating suffering is a just goal, it is crucial to consider solutions that build compassion as well. I argue that many of these solutions should be human, even if AI tempts us with machine speed and efficiency.

If we turn to machines too rapidly and absolutely, we are in danger of what I call dehumanization: the process by which individuals prefer digital lives and interactions to human ones.

Studies show that treating loneliness with AI companions can cause people to replace human relationship with digital ones. Researchers are already strongly recommending against sending teens to AI for mental health support. Therapists are raising the alarm that going to AI instead of human therapists can cause emotional dependence and increased anxiety.

Automation risks causing humans to turn away from connecting with each other and toward insular experiences mediated by computers. The consequences are already arriving in the form of AI psychosis, and the results are dire.

Social isolation can be as dangerous as smoking 15 cigarettes a day, but synthetic connection can deepen the problem.

Finding the Automation Boundary

We are quickly reaching a point with AI automation (as we are in so many AI-related fields) where “can we” is not a sufficient question. My heuristic provides the next five:

1. Does this enhance human agency?

2. Does it deepen trust and connection?

3. Does it sharpen human judgment?

4. Does it expand collective opportunity?

5. Can its outcomes be audited or reversed?

By accurately answering these questions to determine the automation boundary, we can eliminate vicious friction, preserve virtuous friction, and realize the responsibility of directing our increased time, energy, and awareness (as provided by AI) towards solving those problems that demand, and deserve, human solutions.

Visit zackkass.com to learn more about Zack and get in touch.

AI has already had some bizarre downstream cultural consequences. The average “corporate influencer” now has a fiduciary obligation to bill themselves as an AI expert — they’re often rewriting their past to do so — and *everyone* has the tools to “sound” smart by filling their newsletters and social media posts with big (meaningless) words. Syllable count has replaced concept quality as a means of communicating value. Exchanging ideas has become a game of word vomit performance art.

The result is AI slop (often about AI slop).

The worst part: algorithms are rewarding this editorial potpourri, elevating “experts” who trade on “smart-sounding” slop, publishing tens of thousands of words that somehow say nothing of real substance, and still attract enormous audiences. It exacerbates the entire problem. I don’t want to read it — I genuinely hate it; and I think most people do, too — and we certainly don’t want this trash shaping the zeitgeist for a few reasons:

1. Cultural malaise and “copy-pasta” aren’t good for the soul — McMansions don’t produce beautiful neighborhoods, etc.

2. Important ideas get ignored in favor of whatever drives attention — likes, views, impression.

3. And most importantly: polished-but-empty language conditions us to disregard what words actually mean.

Normalizing imprecise language is making (a lot of important) conversations about complex topics (like AI) overly simplistic and sloppy — and I see it as my job/purpose to bring nuance/precision to some of these issues.

As it relates to this blog post, a particularly glaring example of this is the indifference with which most people use the terms “knowledge” and “intelligence.”

Knowledge vs. Intelligence

Knowledge is acquired and stored information. It is all the facts, data, skills, experiences, know-how, and patterns we accumulate. Whether in humans or machines, it does not act on its own. It is an asset, a resource.

Intelligence is the capacity to interpret, transform, and extend knowledge — to glean meaning. Intelligence allows us to reason, learn, adapt, abstract, and make new connections. It is the tool we use to answer questions, solve problems, and seek deeper and broader knowledge.

Knowledge is raw power. Intelligence is the ability to use that power to do work.

Confusing the two causes people to misunderstand the scope/consequences of the internet (unmetered knowledge) and AI at scale (unmetered intelligence).

To realize the full potential of unmetered intelligence — limitless cognitive capacity delivered by AI at near-zero cost — we need to understand the difference between knowledge and intelligence, and how dramatically AI is going to change our relationship with both.

The Evolution of Knowledge

For most of human history, knowledge was scarce and inaccessible. In the 1700s, even Harvard’s library held only a modest 3,500 volumes. By the 1800s, thanks to the continued spread of Gutenberg’s printing press, an individual may have owned a few books and, if they were lucky, had access to a private church or society library.

By the 1900s, the explosion of public libraries in America expanded access to around 3,500 public libraries with tens of thousands of books in each. By the latter half of the century, it was not uncommon for even medium-sized libraries to have over 100,000 volumes, while Harvard’s ballooned to over 10 million.

This linear growth in access coincided with technological and societal advancements, but it pales in comparison to the logarithmic explosion of access created by the arrival of the internet.

In a single generation, we went from gatekept and geographically bound information to a world where almost every fact and framework is just one search away. (And that, folks, is why we call it the information age.)

By the start of 2025, the internet had an estimated 3.98 billion pages. Wikipedia has 64 million pages in English, and hundreds of millions of data files. Google Books hosts over 40 million books in over 500 languages, and the Directory of Open Access Journals has more than 22,000 open access journals and nearly 12 million research articles.

Across the planet, 5.35 billion people use the internet to access knowledge; that’s roughly 66% of all humans on Earth. In America the percentage jumps to 95%. Google handles 5 trillion searches a year. Online courses have boomed, with Coursera alone listing more than 142 million users. Even YouTube, for all its cat videos and MrBeast challenges, sees 70–86% of visitors using it to “learn new things.”

Humans today have more information at our fingertips than the greatest minds of history, Galileo, Newton, or even Einstein, could have ever hoped for. The tragedy is that it arrived before our human operating systems had a chance to keep up.

We are trying to process unmetered knowledge — frictionless, instantaneous access to the sum total of the world’s information — with brains only a few evolutionary steps from apes. We are filled to the brim, overflowing with knowledge, and it haunts us.

This tragedy plays out in two distinct ways. First, we are more aware of the goings-on in the world than any one of us reasonably should be. Any seven-year-old has access to the Darfur genocide, news reports of every single local murder, and the emotional devastation of losing a partner in Letter to Yi Ŭngt’ae from 1586 (”We’ll be together until our hair turns gray, then die together, so how could you go and leave without me?”).

There is value in knowing, but we are navigating a RAG marketplace with GPT-2 minds. Our brains run at human speeds, and are being served knowledge at machine pace. We are telling everyone, everything. We feel suffering at global scales. We cannot provide context. We can’t efficiently switch between topics. We can’t take a step back and see that, in totality, life on Earth continues to get better.

Second, we are not capable of mining the value of our knowledge for scientific discovery and progress. It is not unreasonable to think that somewhere across these pages, buried in millions of cells of data, we have the essential knowledge to solve cancer, fusion energy, or deep space travel.

We are buried in riches of data, and a Google search is not enough to dig us out. Unmetered intelligence is.

The Boom of Intelligence

In much the same way that the internet changed our relationship to knowledge, AI is scaling our access to intelligence toward unmetered intelligence.

First, models are already reaching human-level intelligence. By 2023, ChatGPT-4 showed “human-level performance” on professional and academic exams, including scoring near the top 10% of test-takers on a simulated bar exam. GPT-5 scored 94.6% on the 2025 American Invitational Mathematics Examination, a test of complex math and problem-solving questions; the median human competitor scored 27–40%.

In 2024, Gemini Ultra became the first model to beat the average human score on the Massive Multitask Language Understanding (MMLU) (89.8% vs. 90.04%), the go-to reasoning benchmark with graduate-level tasks from 57 topics including law, ethics, and history.

In the AI Index 2025 report, the Stanford Institute for Human‑Centered Artificial Intelligence reported that intelligence is only accelerating. In a single year, model performance improved 18.8% on MMLU, as well as 48.9% on the Graduate-Level Google-Proof Q&A Benchmark (GPQA), another reasoning test. Models also improved an extraordinary 67.3% on SWE-bench, a benchmark that requires models to read live GitHub projects and generate effective patches.

These benchmarks require deep reasoning and manipulating knowledge; outcomes that don’t come from a search function or scanning databases.

Models are improving their capabilities so rapidly that they are outpacing our ability to create benchmarks to test them. The frontier is being pushed not by expanding what AI knows, but by expanding how well it can think.

Second, and what makes me most excited, AI is creating intelligence at scales far beyond human capacity.

In September 2025, OpenRouter reported that ChatGPT generated 78.3 billion tokens in a single day. Think of a token as the basic ‘currency’ of language models — a tiny chunk of text (like a short word or piece of a longer word) that the model reasons over.

One token is about four characters, and 100 tokens is about 75 words. If an exceptionally fast human read 300 words a minute (the average adult can read about 238), we could expect them to process about 400 tokens a minute, and 576,000 tokens a day.

It would take that unsleeping, undying human 370 years of nonstop reading to match the cognitive work of ChatGPT from a single day. And as models improve and hardware becomes more efficient, available AI intelligence expands.

Unmetered intelligence addresses both tragedies of unmetered knowledge. First, it gives us our first meaningful tool to fight against the wave of knowledge that has overwhelmed us. AI can act as a membrane, synthesizing immense quantities of knowledge and only letting the necessary takeaways and context through (for example, GPT-5.1 provided a version of this blog for 12-year-olds and a version of this blog for 7-year-olds, but people of any age can use these tools to parse knowledge more effectively).

Second, it can add cognitive power to fields long limited by scarce human expertise — oncology, astronomy, nuclear physics, and more — to process existing knowledge toward new insights and discoveries.

The biggest impact of AI will come not from it being more intelligent than us (though in some arenas it already is, and in many others I think it will be soon), but rather from having more intelligence than us.

Intelligence in Action

After more than 50 years of ingenuity, rigorous science, and specialized human intelligence, the Protein Data Bank — a public, global database of protein structures — had identified about 190,000 unique protein structures.

In 2022, Google DeepMind launched the AlphaFold Protein Structure Database to do much the same work. In just three years, the open-source project used AI to predict over 200 million protein structures.

AI used the same knowledge available to the human scientists, but applied intelligence at scales large enough to solve the problem.

Perhaps all of the scientists in the world, with unlimited time, could have found each of these protein structures (I doubt it), but it would have taken generations and perhaps the total focus of humanity.

190,000 in 50 years vs. 200,000,000 in 3 years. The work earned the team behind the AlphaFold Protein Structure Database a Nobel Prize.

The Shift

The scope of human knowledge has long been confined to the answers of “definite problems.” How far away is the Sun? How large is the Earth? What are molecules made of? How can we improve crop efficiency?

But unmetered intelligence gives me hope that we will soon have enough cognitive capacity to aim for even larger questions. Questions that human intelligence could never comprehend, “indefinite problems.”

What is our purpose? What does the fourth dimension look like? How might alien life forms exist beyond our anthropomorphized imaginations?

There is a clear divide between intelligence and knowledge, but AI is increasing our access to, and the fruits of, both.

Visit zackkass.com to learn more about Zack and get in touch.

Last week I introduced The Adaptability Trap — when the instinct to change overwhelms the instinct to lead — and how to avoid it by honoring core beliefs. (If you haven’t read Part I, I recommend doing so before reading this Part II).

Implicit in this framing is an obvious question: How do we identify our core beliefs and differentiate what to change from what to protect?

To make sense of this, I’ll tell you a story my hippie uncle told me (with much greater detail when I was far too young).

On the evening of July 3, 1973, Ziggy Stardust stood on stage at the Hammersmith Odeon in London for a concert that was the culmination of a more than yearlong, three-continent tour to promote The Rise and Fall of Ziggy Stardust and the Spiders from Mars.

As the show reached its climax, Stardust stood at the mic beneath a shock of crimson hair and announced, “Not only is this the last show of the tour, but it’s the last show that we’ll ever do.” In front of the stunned crowd, the band played their final song: “Rock ‘n’ Roll Suicide.”

The androgynous, interstellar “alien rock messiah” who had risen to global fame was no more. All that remained was the man beneath the makeup, David Bowie.

The persona of Ziggy Stardust had grown large enough to endanger Bowie’s real purpose: personal expression, creative sovereignty, and intimacy at scale. In ending Ziggy Stardust at the apparent zenith of fame, Bowie demonstrated his ruthlessness in separating his purpose — why he did things — from his ways and means — how he did them.

While it may have been difficult, by “killing” Ziggy Stardust Bowie preserved his core, created a path toward five decades of musical success, and avoided what I call The Ways and Means Trap: when the instinct to do the same thing overwhelms the instinct to do the right thing.

The Accelerating Pace of Change

As I wrote a few weeks ago, the costs to use AI (what we call inference costs) are plummeting; per-token prices have collapsed and ultra-low-cost models are becoming mainstream. On top of this, model performance continues to improve. Kimi K2 Thinking, an open-source model, now competes with, and in some instances exceeds, the capabilities of ChatGPT-5 and Claude Sonnet 4.5. AI is making experimentation and failure essential and accelerating the pace of change.

Last month, OpenAI reported ChatGPT had 800 million weekly active users, nearly twice as many as in February of this year. A recent McKinsey study shows that 88% of organizations now report regular AI use in one or more business functions, a growth of 10% from 2024. More than half of respondents reported experimenting with AI agents, and nearly a quarter are scaling agentic systems.

These are not small shifts or pilots; they are transformational endeavors. 21% of companies say they are using AI to fundamentally redesign workflows, and among “AI high performers,” about 50% intend for AI to transform rather than optimize their business. A global IBM study showed that executives expect AI-enabled workflows to grow from 3% to 25% by the end of the year. Already, 64% of AI budgets are spent on core business functions rather than experiments.

Across industries, AI is changing how work at the individual and structural level is done.

The Place for Purpose

Amidst this change, companies are increasingly talking about, and pointing to their purpose as a means of recruiting, press, and even growth. A corporate purpose serves (performatively or otherwise) as the company’s North Star for decisions regardless of circumstances or challenges. The definition of purpose can broadly be expanded to include mission, vision, and values.

Delta Air Lines’ mission is “to connect the world.” Biotech company Illumina’s vision is “to provide access to genomic medicine for every patient in need.” Nike’s mission is “to bring inspiration and innovation to every athlete in the world.” Walmart’s purpose is “saving people money so they can live better,” while UPS says theirs is “moving our world forward by delivering what matters.”

Whereas purpose is the why, ways and means are the how: everything from strategies, people, processes, tactics, and tools.

We are entering a chapter in which, I think, a K-curve will emerge as companies sort by those committed to their purpose vs. those committed to their ways and means.

I won’t over-editorialize a familiar story, but in 2002, while owning 90% of the non-theater entertainment market, Blockbuster’s mission statement was “to be the global leader in rentable home entertainment by providing outstanding service, selection, convenience and value“. They had confused their purpose with their ways and means. By 2007, they had lost almost all of their market share to Netflix which realized that DVDs and return boxes were discardable in the pursuit of their purpose, “entertaining the world”.

Today’s professional services organizations are at a similar inflection point. So much has been made of the demise of management consulting, for example. Quite candidly, I think the bet against these companies might be a good one, but not because AI is obviously going to eliminate the need for strategy consulting. I think the bet against these companies is fair because many of them will prove to be rigidly committed to their ways and means.

Despite mission statements like “to help our [McKinsey’s] clients make distinctive, lasting, and substantial improvements in their performance,” I think the partnership model (in which most of these firms are organized) will prevent them from investing in products and systems that would provide super-linear efficiencies to accomplish their mission. I think some of the famous consulting firms will rent associates on a per-hour basis until the last client signs a contract, in much the same way that the last Blockbuster still offers DVDs to the handful of families in Bend, Oregon, who want to rent them.

Upstart and adaptable professional services companies will maneuver past the legacy players by combining AI and human services to offer more innovative products and bend pricing models to provide step-function improvements in value and economics. Their missions will sound the same; their ways and means will be very different.

The Ways and Means Trap affects more than just companies, though; individuals experience it as well. In each of our lives, professionally and personally, it is crucial to identify our own purpose.

This could be “to be a teacher” (Oprah Winfrey), “making people happy” (Walt Disney), ”becoming the heavyweight champion of the world” (Lennox Lewis), “being funny, on my own, in a room full of strangers” (Jerry Seinfeld), or any other number of noble personal pursuits.

When individuals identify purpose, they, like companies, can continuously develop the ways and means to best pursue it, and adapt them through waves of change.

To everyone reading this, if your identity feels threatened by technology, you may have mistaken your ways and means for your purpose, and fallen victim to the Trap. To work your way out, try to identify what your purpose is, and then consider how technology can help you amplify it.

Doing, Not Saying

The modern corporation is eager to proudly state its purpose, mission, vision, and values. But CEOs rarely get hired and fired for their commitment to the mission. When it comes to making tough decisions, capital returns often take precedence.

Companies are often bound to established and increasingly inefficient ways and means that prevent directional adaptation. Actual commitment to, and investment in, purpose allows leaders to experiment.

Individuals and businesses can test theories, do deep research, pivot from negative results, and explore improvements without exorbitant costs or company-wide buy-in. These experiments, across departments, industries, employees, and methods, become more accurate and actionable as they can all be judged against the same criteria: Does this advance our purpose?

This opens the door for leaders and companies to broaden the scope of how they operate and serve customers, including entering adjacent industries, evolving workflows, and developing entire new verticals of products and services.

How might “connecting the world” look when leaders are able to consider new modes of transportation beyond airplanes? How could “access to genomic medicine” be better realized with AI-enabled experiments and medicine development? How could a company better “save people money” when supply chain models, product suggestions, and delivery methods can all be reconsidered? So long as companies follow the through line — advancing purpose — the vectors of productive opportunity widen.

The Freedom of Purpose

When leaders anchor the why and interrogate the how, they unlock Bowie’s kind of freedom — the confidence to retire past success, protect the work that matters, and continuously experiment. Individuals and companies that do so turn Behavioral Adaptability into an advantage, and avoid both the Adaptability Trap and Ways and Means Trap.

Employees gain real agency and permissionless autonomy to adapt toward purpose. Agreeableness becomes obvious and incompatible with the work required to stay current and constantly improve. Teams can debate, run cheap experiments, abandon old methods, express differing views, and compound progress without losing themselves.

In an age of Unmetered Intelligence, this is how to make an advantage out of uncertainty. Abandon nostalgia, redesign systems toward the vision, let every change pull you closer to the mission, and at all times align yourself with authentic values.

Protect the purpose. Evolve the rest.

A few weeks ago, at LinkedIn’s AI in Work Day event, chief economist Karin Kimbrough stood before the crowd and confidently declared that in a changing world, “adaptability is the new currency.”

As definitive statements go, it’s as banal as it is aspirational.

Adaptability has always been a virtue, and it is one of the critical (and unique) features of Homo sapiens’ evolution. But adaptability carries a hidden pitfall, especially during times of radical change.

When the drive to adapt overwhelms everything else, it can hollow out conviction. History has watched spineless leaders twist themselves into whatever shape they think the moment calls for, forgetting that some tides are supposed to be resisted, and not all change merits adaptation.

We love to tell stories of companies that failed because they didn’t adapt — Blockbuster, Kodak, BlackBerry. But just as many have died from over-adapting. WeWork reinvented itself until no one knew what it was. Yahoo chased every trend until it stood for nothing. GE pivoted so hard it forgot what business it was in. Even Nokia, once the world’s most innovative handset maker, collapsed under the weight of too many changes.

These weren’t failures of imagination; they were failures of conviction.

This, to me, is the Adaptability Trap: when the instinct to survive or impress overwhelms the one to lead. Change becomes more appealing than core values, and purpose is lost in the shuffle.

What’s the Buzz on Adaptability?

“Adaptability” has become a catch-all strategy for an unnervingly unclear near-future for two reasons.

The first is that it’s aspirational and sells LinkedIn feel-goodery slop. It lets individuals feel like control is still in their hands; they just have to change enough! The second is that it allows leaders to feel like they are building an arsenal to solve any problem (and convince the powers above them that they are as well). If they just hire endlessly adaptable workers, the company will change to meet whatever the future brings. Problem solved.

Scale AI’s interim CEO Jason Droege recently told Business Insider, “The world’s changing, right? So you do need people that are adaptable.” In a late 2024 SAP corporate blog, the company stated that shifting skill demand in the workplace “underscores the need for adaptability as workers learn to stay relevant in an evolving job market.” They reinforced this position in early 2025 saying organizations can “create a people ecosystem centered on adaptability and growth.”

The U.S. Bureau of Labor Statistics now includes “adaptability” in its Employment Projections skills taxonomy, formalizing it in skills frameworks. In the World Economic Forum’s Future of Jobs Report 2025, 67% of global employers labeled resilience/continuous learning — foundations of adaptability — as rising core skills, an increase of 17 percentage points since 2023, leading the report to single out “the critical role of adaptability.” These skills only lagged behind “analytical thinking” in skills rankings.

Adaptability is ascending for good reason. But I fear that — like the child who eagerly runs to the top of the snowy hill only to realize they forgot to bring their sled — our pursuit of adaptability is leading us to leave the most important things behind.

The Chameleon and the Shapeshifter

It’s a bit simplistic, but I think there are basically two archetypes of adaptation: the Chameleon and the Shapeshifter.

Chameleons can fit most anywhere while staying true to who they are. They shift tone, language, and behavior to match their surroundings, but their core beliefs remain intact. They read the room well, and know who they are when they leave it. They’re easy to get along with, but their adaptability has limits. The color changes; the creature doesn’t. In moderation, this kind of adaptability is a strength — it’s insight and empathy in motion. Plus, they’re fun to be around.

Shapeshifters, on the other hand, will abandon everything to try to become the “perfect” fit. They don’t just change behavior; they change form. Their flexibility is absolute — a full-body surrender to circumstance. Think of the man who, under the full moon, undergoes grotesque transformation into a werewolf. Nothing remains of the old self, no structures, no beliefs, no morals. The Shapeshifter forgoes the foundations of the last form in pursuit of the benefits of the next. They’ll say whatever gains approval, endorse whatever wins favor, and they mistake malleability for growth. They’re unnerving people to be around.

Both thrive in systems that value adaptability above all else — they’re masters of change. But when organizations reward the ability to contort and change more than the courage to stand firm, they start selecting for Shapeshifters — leaders who can survive any challenge or change, except the test of conviction.

The Cog and the Monomath

Though they may be falling out of favor, understanding the archetypes of rigidity, the Cog and the Monomath, can help us better understand the Chameleon, the Shapeshifter, and the Adaptability Trap.

Cogs do not adapt, and they do not stand up for themselves. These individuals simply do their job. Anything beyond function is beyond the purview of the Cog. Core values are inconsequential compared to the task they are assigned. Think of the ditch digger, they simply dig ditches.

The Monomath is rigid in their pursuit, and firm in their convictions. Their area is that of singular focus and deep expertise. Their core values, principles, and mission remain unchanged as they constantly work to refine their craft, to zoom in even further. Think of a 100-meter runner. They are focused on running, but they work to get faster.

Agency and Agreeableness

While these archetypes seem opposed, they snap into focus when we consider them across the axis of agency. As I recently wrote, agency is the common corollary to personal achievement.

Adaptability allows us to change; agency allows us to stand firm, to question, and to grow.

The Chameleon and the Monomath express agency. They preserve what is important to them and stand by their convictions while deepening expertise and abilities. How they go about pursuing the things that matter to them differs (adaptability vs. rigidity), but they share the power to do so. By leveraging agency, they move forward while holding their core values close.

Opposite agency is a trait that absolutely drives me wild: agreeableness. The Cog and the Shapeshifter are infinitely agreeable. They do not stand firm to any core beliefs or convictions. The rigid Cog has no core beliefs to hold on to. The adaptable Shapeshifter is willing to change any core belief in order to survive and impress.

Without guardrails — vision, mission, values — adaptability becomes moral drift. It reminds me of a passage from Lewis Carroll’s Alice’s Adventures in Wonderland:

ALICE: Would you tell me, please, which way I ought to go from here?

CHESHIRE CAT: That depends a good deal on where you want to get to.

ALICE: I don’t much care where.

CHESHIRE CAT: Then it doesn’t matter which way you go.

When the destination is undefined — the purpose for the journey — any pivot feels justified, and change feels like progress.

The goal isn’t to eliminate adaptability but to anchor it. Be Chameleon enough to understand and adjust to the world around you, but not such a Shapeshifter that you lose yourself within it. Build environments where employees can express their own agency, while preserving the company mission and values. The challenge of leadership in an age of constant change is knowing what must evolve — and what must endure.

The Dangers of Infinite Adaptability

Within the Adaptability Trap, low-agency employees learn there is no reason to push back, and no vision to push for. This creates workplaces full of sycophants whose easiest path forward is to nod, agree, and react.

Employees “figure it out” rather than finding ways to “make it right.” A worker beaten down by a flawed internal system, a poor manager, or supply-chain inefficiency just deals with it. They don’t hold firm to company beliefs, challenge assumed outcomes, flag concerns to management, or warn of values drift.

Research shows that when people tailor how they present themselves to every situation, or “self-monitor,” they exhibit lower attitude‑behavior consistency. What they say is less likely to predict how they will act. When hiring for unlimited adaptability, employers risk getting flaky, unpredictable employees.

For individuals, the Adaptability Trap is a race to the (moral, financial, and vibe) bottom. It promises agency but actually strips people of it. Why would an employee flag an issue if they felt it would signal to their employer that they aren’t adaptable enough? Infinitely adaptable people become agreeable to anything and accountable for all of it. They will say yes to ever-expanding KPIs, absorb “just one more” project, forgo pride in work, pursue completion, and abandon the mission that may have attracted them to the organization in the first place. They will adopt all the tools, and get real value from none of them.

The dangers seep all the way into personal identity. Research suggests that constant personal adaptation can cause the image of yourself, the self-concept clarity (SCC), to become fuzzy. When we change for everything, we start to forget who we are. Low SCC predicts stress, depressive symptoms, lower self‑esteem, neuroticism, greater susceptibility to external cues, poorer self‑regulation, and poorer life satisfaction.

There’s Never Been a Better Time to Disagree

What frustrates me most about the current worship of adaptability is how misplaced it’s become. We keep telling people to bend, pivot, and adjust — but rarely to question. The irony is that there has never been a better time to express your agency, to hold your ground, to test assumptions, to challenge the direction of change.

For most of history, agreeing and adapting made sense. Your boss, your teacher, your elders did know more than you. They’d seen more of the world, faced more challenges, and had access to information you couldn’t dream of. Adaptability and agreeableness were survival.

But that’s no longer true. The monopoly on knowledge is gone. Anyone can access the world’s ideas, data, and expertise in seconds. If your boss makes a claim or a decision that feels wrong — or betrays the mission — you no longer have to nod in silence. You can test the logic, pull the evidence, and model the outcome yourself.

Adaptability without discernment is regression.

Adapt How, Not Why

Avoiding the Adaptability Trap, while staying relevant, requires exercising agency and conviction alongside adaptability. By preserving those things most important to their core — values, mission, vision — leaders and companies can lead while meeting the evolving demands of the outside world.

This is the promise of what I call Behavioral Adaptability — the capacity to alter one’s practices and habits in response to changing conditions, while maintaining one’s deeper principles. This creates stability through change, and builds a foundation that allows employees to question, learn, and grow.

There is a clear distinction here from the Adaptability Trap. The Adaptability Trap is a willingness to change everything, to be a Shapeshifter. Behavioral Adaptability is a willingness to change how you pursue what is most important to you, to be a Chameleon. Agency begets agency, and builds safe sandboxes for experimentation and progress.

Leaders and individuals need to determine when to adapt, where to push back, what to change, and what to never surrender. This is the way to proactively prepare for a changing future, without leaving yourself behind.

How to do so — is a topic for next week.

The MIT report (the one that found 95% of business AI pilots fail) made waves over the last month in business circles, and beyond. Journalists pointed to it as proof of a bubble, nervous executives promoted it as evidence that AI investment is massively wasteful, and practitioners seized on the moment to remind their bosses and peers how impressive their success is.

All of these interpretations (albeit reasonable) miss a major point that I’ve explained to execs for the last month.

95% failure doesn’t mean anything if we don’t consider what it costs to fail. And a 5% success rate doesn’t mean anything if we don’t consider what it pays to succeed.

Moreover, even MIT isn’t saying to not use AI at work — they’re selling courses for more than $20,000 aimed at executives to harness AI for business success.

Plainly, I’m not concerned with the 95% failure rate at all. In fact, I don’t think we are failing enough. Historically, the constraint on experimentation has always been cost — but that constraint is vanishing.

The Falling Cost of Failure

Adolphe Sax was obsessed with experimentation. In pursuit of a better sound, the 19th-century instrument maker constantly changed materials, key layouts, sizes, iterating over and over again, combining elements of woodwinds and brass until he found things he liked.

First, these experiments led him to improve the bass clarinet. Then, he built an entirely new family of instruments, the saxhorn, followed by the saxotromba, before finally creating the instrument that would put his name in the annals of brass, rock, and jazz: the saxophone.

Experimentation and iteration allowed Sax to explore his ideas, iterate on success, and create something new. It is also one of the fastest paths to find new ideas, solve unsolved problems, rethink past solutions, and discover the next problems to solve.

The thing about experimentation, though, is it can be prohibitively expensive. The Human Genome Project is estimated to have cost around $3 billion. The DOD and HHS obligated $13 billion to Operation Warp Speed, the effort to develop the COVID-19 vaccine. Adolphe Sax went bankrupt three times and had to sell his entire personal collection of instruments before dying of modest means.

When even success is expensive, the cost of failure, even the threat of failure, can stifle innovation. Though failure is a feature of experimentation, it has long been one of its largest constraints.

For many years, this has been the reality of AI. Training GPT-4 cost an estimated $80–100 million. Meta put the number for training Llama 3.1 at hundreds of millions of dollars. Goldman Sachs estimates that “The five highest-spending US hyperscale enterprises will have a combined $736 billion of capital expenditures in 2025 and 2026.”

Crucially, though, while training a frontier model requires resources comparable to those of a small nation, the cost of using a model, the inference cost, has been in freefall. From 2023 to 2024, the price per token between GPT-4 and GPT-4o fell 150x. Sam Altman projected that “the cost to use a given level of AI falls about 10x every 12 months.”

Every year, it is getting cheaper to access the intelligence of AI. We are approaching Unmetered Intelligence, when intellectual capacity becomes as ordinary, accessible, and dependable as flipping a light switch. Compute, tooling, and algorithms are becoming as inexpensive to use as the electricity required to run them.

This means that any of us now have the capability to wield this intelligence. And when using a model is cheap, the cost of failure becomes cheap. The barrier to experimentation is disappearing. And while the experiments above were undertaken in specific fields, experimenting in AI can help us gain insights across industries and curiosities. This widens the scope of wonder, and the potential for discovery. We can do more, we can do it faster, and we can iterate at scale.

How Did It Get So Cheap?

To understand how exactly inference costs have gotten so low, and why they’ll continue to decline, it helps to get in the weeds a bit on the economics of compute.

The largest set of reasons to point to is the same reason you pay less for gas in a Civic than a Silverado: efficiency. This efficiency plays out in a number of ways. We are getting more efficient at physically manufacturing compute power: the compute power that cost $18.7 million to build in 1984 cost $0.03 in 2017. This is known as Wright’s law. The computers themselves are also requiring less energy to do the same work; a recent report found that compute energy efficiency doubles every 2.29 years. This is Koomey’s law.

The AI algorithms themselves are also getting more efficient; in January 2025, DeepSeek released its DeepSeek-R1, which achieved performance comparable to other models while being 95–98% cheaper per task. And while demand for AI performance continues to increase, the efficiency of the data centers behind them continues to improve through scaling and improved infrastructure.

Innovations and competition are also decreasing costs. Increasingly, models are batching requests so that GPUs spend less time idle waiting for requests, and, of course, normal market factors (competition) are causing costs of cloud compute to decline, offering more, and more cost-effective options to run models.

How to Experiment More: The r-Selected Species

With these cratering costs (which look like they’ll continue to fall) the experiment tax is collapsing — when the cost of failure is low, the cost of many failures is only marginally more. That flips the calculus of exploration and failure: the downside is minimal, and the upside, especially at scale, can be tremendous. The advantage shifts to those who convert that deflation into more shots, sooner, with stronger learning loops.

This is going to require change. We don’t like failure, and many of us have a hard time learning from it (at least in the short term). Success requires a mindset shift. It is time to channel our inner r-selected species.

You might recognize the term from high school biology: r/K selection theory. It separates species into two groups: r-selected and K-selected.

Humans, biologically, are a K-selected species. Our strategy to raise our young successfully, so that they can go on to raise their own young, is to invest in them heavily. We can only have a few offspring (no litters), they take longer to reach maturity, and they need heavy parental care. If it goes well, our offspring are larger, they live longer, and they have a higher chance of survival. K-selected species thrive in stable environments. We’re not the only species that does it, so too do elephants, whales, and emperor penguins. It’s a quality over quantity choice.

r-selected species take a different route. Theirs could be described as quantity over quality. They have many offspring, they’re often smaller, they mature early and require little parental care to go off on their own. Many will die, but many will also survive. These aren’t really species that we often laud (I was going to say lionize, but — K-selected species), they’re the ones that thrive in instability: mice, fruitflies, bacteria.

To be successful in the future, you must run experiments like fruit flies: cheap, fast, and plentiful. Experiments can spread far, spread fast, investigate any opportunity, and swarm on success; building on one another in wave after wave of experimentation.

The Individual and Company Swarm

For companies, a culture of experimentation should not be considered aspirational, it needs to be treated as forward-compatible infrastructure. It is not enough to run quarterly “innovation sprints.” Companies must become perpetual idea and implementation factories: experiment with copilots in familiar apps, AI assistants, calendar managers, email triage, client outreach, customer service, and look for any other advantage (or more stars like the student I wrote about last week).

Rather than going for K-selected “big bets,” capital-intensive gambles that require extensive compute and human brain power, try micro-bets, and try a lot of them. Hypotheses are no longer budget decisions, they’re keystrokes. Your goal isn’t one big win; it’s a steady drumbeat of small, cheap trials that compound into advantage.

This isn’t only true of businesses though, each of us, personally and professionally, can experiment to achieve and ship more in our lives. Treat more questions, tasks, and ideas as opportunities for experiment: draft the memo, prototype the feature, spin up the side project, test ten versions, keep the ones that work and build on them.

Do this because it’s interesting, and do it because automating rote tasks and expanding your mind is a moral good. Find ways to do more and reclaim hours of your life. More and more I’m meeting people who tell me experimentation with AI has led them to building systems that allow more time for everything else. The more we explore with AI, the less we have to actively use it.

Quality ideas, care, and motivation will still matter, but we can direct them more efficiently by leveraging quantity and speed. Experimentation isn’t a detour from progress; it’s the shortest path to it.

Fail Effectively

As you experiment and embrace failure, make sure you do so while allowing yourself to succeed. Measurement is the difference between motion and progress: know exactly what you’re testing, and learn from the result — success or failure — so you can act. The goal isn’t to “win” every experiment; it’s to convert ambiguity into learning you can use.

Frequent failure can be emotionally expensive, so normalize it. Highlight what was ruled out as progress, and separate self-worth from hypothesis worth. Prepare for legacy resistance by creating time, space, and safety to fail toward progress without raising a red flag for higher-ups. And in high-risk domains — user impact, data sensitivity, regulatory exposure — don’t just experiment fast; experiment safely with human-in-the-loop checkpoints, reviews, and rollback plans. Speed where it’s safe, and safeguards where it’s not.

Build, Build, Build

As our capabilities continue to surge, and our needs and interfaces keep shifting, develop the muscle to reinvent frequently. Unmetered intelligence means that experimentation doesn’t end at financial ruin, and it doesn’t end at success; it’s a pathway for sustained progress. We have to keep walking it.

So my call to you is to experiment, and build. And do it frequently.

The only rational response is to try more, and the only real cost is in not failing enough.

Intellectual decline, especially with the arrival of AI, has been a major topic of concern lately; 61% of parents think greater AI use will harm students’ critical thinking, 55% of high schoolers agree.

I’ve written and talked about this issue quite a bit. Some of you have heard me discuss my theory of “idiocracy.” It describes the state an individual reaches when they decide that critical thinking is inconsequential to survival, and stop engaging in it.

Are We Getting Dumber?

When I talk about idiocracy, it’s often met with a chorus of disheartened nods. Most parents I speak to are worried about the general population getting dumber. This concern is certainly not without merit.

The NAEP 2024 (America’s education report card) shows only 28% of eighth graders are proficient in math, down 5% since 2019, and only 30% are proficient in reading, down 2% from 2019 and 5% from 2017. In low- and middle-income countries 70% of ten-year-olds are unable to read a simple text with comprehension, ballooning from 57% in 2015. The decline can be seen physically as well, 77% of pre-k to third grade teachers say students have more difficulty using scissors, crayons, pencils and pens than five years ago; early exposure to screentime is associated with fine-motor delay. Fewer kids can swim and ride a bike, per capita, than 20 years ago. All told, GenZ and GenAlpha, on average, appear to be the first generations in many to not be smarter than the prior—on average.

First, I should say: it’s pretty clear this condition predates AI. ChatGPT serves as a convenient scapegoat for the undoing of 20 years of education progress in literacy, but the evidence suggests we began this decline during the COVID lockdown (coinciding with the erosion of in-person education and a dramatic rise in screen time among children).

Second, it’s not limited to young people. In 2023, 28% of US adults were low performing in literacy, an increase of 9% from 2017, and 34% were low performing in numeracy, an increase of 5% from 2017. More and more adults are choosing to “switch off,” adult internet users around the world average 6 hours and 38 minutes online each day, with 31% of American adult social media users self-reporting daily doom scrolling. We now live in a world where is is possible to stare at your device for 16 hours a day, order DoorDash for every meal, and effectively turn off your brain.

But…buried in this data (because good news doesn’t sell ads) is a far more nuanced and promising story. While most parents I meet are worried about other kids, they often regard their child as much smarter than they were, at that age; and the data suggest that it’s not just their pride speaking.

The Rising Ceiling

While the average student appears to be in a recent decline, many are doing better than ever before.

More than 2.8 million US high schoolers are taking college level courses, (fig. 1) and the number has grown more than 10% year-over-year. 35.7% of 2024 public-school grads took an AP exam, up nearly 3% from 2014. Students are earning the State Seal of Biliteracy at increasing rates, with double digits gains in some areas, including a 29% growth in New York City. Around the world, more than 96,000 students are enrolled in the IB Diploma Programme, an increase of more than 5% between 2023 and 2024.

Moreover, this performance cuts across socio-economic, gender, and geographic demographics. More than a quarter of AP exam takers in 2025 were low-income students; female enrollment in AP Computer Science is up 81% since 2017; IB programs are expanding in Ghana, Botswana, Togo, Peru, and beyond.

Young people are looking for opportunities to push and test themselves outside the classroom as well. Around the world, participation in FIRST robotics, a “sport-for-the-mind” robotics program, reached more than 785,000 students. Regeneron ISEF 2024, the world’s largest pre-college STEM competition (think: a super science fair) hosted 1,699 finalists from 67 countries and regions. 24% of 18–24-year-olds are actively starting or running a business.

Success, at an expert level, is arriving earlier and often outclassing the establishment.

• A 17-year-old from Spain, won the 2025 Shigeru Kawai International Piano Competition against a field of adults.

• An Argentinian 11-year-old, became the youngest player ever to surpass a 2500 FIDE rating (the international measure of chess strength).

• A 17-year-old student in the UK used AI to remap textbook diagrams for color-blind learners, achieving 99.7% accuracy and expanding access to learning aids.

• An 18-year-old trained an AI model on NASA data, uncovering ~1.5 million previously unknown astronomical objects.

• A Florida high-school senior built an AI system to diagnose dementia with one brain scan.

• An eighth grader created an AI tool to detect dyslexia.

• A pair of 10-year-olds in Nigeria created an app to keep drivers awake behind the wheel.

And the markets are responding accordingly to all of this prodigal success: OpenAI and Palantir (among others) no longer require college degrees for employment.

For the first time ever, education and exceptional achievement are not dependent on access to institutional resources and human experts. Technology has un-gated these pathways to personal development. Entire communities long left out of the mechanisms of upward mobility can finally see pathways to equal opportunity.

Amidst all of the doom and gloom, something very good is obviously happening: technology is expanding potential and opportunity, and giving each of us new control over our own levers of success.

Concern for our general educational well-being is warranted, but we should celebrate that centuries-old barriers to opportunity are crumbling, and new abilities to question, learn, and explore are growing in their place—all over the world.

The Intellectual K-Curve

These two trends combine to form what I see as a new intellectual K-curve (fig. 2), the pronounced divide between those who are in the active pursuit of mental prosperity, and those who passively concede to intellectually rot.

The people who bend up the K-curve are leveraging technology to explore their interests and limits intellectually, creatively, and physically.

The people who bend down the K-curve are often succumbing to technology, passively opting out of intellectual exercise.

Most of those trending down have probably discovered that critical thinking is no longer of material importance to survival. Their opportunities and outlook tend to plummet along with their brain activity.

More and more, individual trajectories on the K-curve are not determined by circumstance, but by mindset.

We are observing, in real-time, that personal agency (or motivation) has become the primary corollary to personal achievement, outpacing even the juggernaut of socio-economic status.

Solving for Agency

Unlike other competitions, mental prosperity isn’t a zero-sum game: no one’s intellectual fulfillment requires anyone else’s failure. As a society, we should work, diligently, to help everyone bend up on the curve. To do so, we need to reward, above much else, agency.

Raising children to see technology as a means to accomplish more rather than a means to do less is one of the most noble things a society can pursue.

To do this, we must acknowledge a few things:

1. There is a cost to providing agency.

2. Our place on the K-curve is not absolute.

3. We can build systems to help more people bend upward.

Machine intelligence is rapidly approaching utility-scale affordability—what I call Unmetered Intelligence—yet what we do with it remains far from commoditized. Unmetered Intelligence does not mean universal brilliance, just as literacy does not mean everyone reads, or the internet means everyone does good research.

Developing this agency is going to be a deeply human pursuit that will require human intervention. Communities without access to guides, mentors, or parents with enough free time to nurture curiosity will inevitably fall behind.

But even those who are down are not necessarily out. The K-curve reflects motion; our places on it are not fixed. This creates opportunity for those on the downward trend to exercise their agency and rise—but it also means that all of us, across our lifetimes, must continuously pursue improvement: learning new tools, adapting methods, and choosing to grow.

So let’s build systems that help people reach the upward bend of the K-curve. Agency is not an innate ability determined at birth; it is a practice that can be taught and strengthened. We can cultivate it in ourselves and in young people. And to learn how, we should look to the kids already exercising it.

First, we need to create arenas to safely exercise agency. This is a key ingredient in many of the success stories above. The teen who discovered space objects did so inside the Regeneron Science Talent Search. The girls who created the driving safety app submitted it to the Technovation Global Summit. The MRI to diagnose dyslexia materialized at a county science fair.

Agency develops in places where failure is a pathway toward greater insights: opportunities like science fairs, Odyssey of the Mind, Lego League, math competitions. These are spaces for students to take risks, explore, develop, and learn outside of the strict architectures of academic achievement—pass/fail, grade point averages, class rankings—that reward completion more than pursuit. Young people need to feel safe to try, fail, iterate, and have their individual successes celebrated.

Second, we need to embrace interest as a pathway to agency. The student who built the color-blindness learning aid did so as a solution to his own struggles with color blindness. The student who designed a single-scan dementia screening had watched his grandfather suffer from the disease.

These young people achieved incredible things by pursuing the challenges and problems they cared about deeply; in doing so, they learned that their capabilities extended as far as their efforts, well beyond the measure of any grading rubric. This creates a sense of possibility, responsibility, and capability to pursue difficult things. Higher-order pursuits instill the confidence necessary to succeed in all tests, not just the one at the end of the school year.

The Pursuit

Ultimately, these strategies teach that what we do matters. Young students feel the consequences of their efforts, actions, and inactions, developing autonomy and ownership. The focus changes from reaching the floor to raising the ceiling—eagerly pursuing the upward bend of the K-curve.

These are humanist pursuits. I experienced them as a child when, even as a “bad” student (traditional schooling and I never quite fit), my parents encouraged my deep interests and gave me room to explore them. But I truly learned their value as a husband.

My wife, Adlee, is a first-grade Waldorf teacher. Whenever I visit her classroom, I watch her focus less on creating supreme test-takers and more on nurturing confidence, resilience, and agency. It’s inspiring to see children realize that they control their own time, attention, and effort—that things happen when they act. Adlee creates more capable humans who grow as they realize that accomplishments, like learning to tie their shoes, are within reach.

Through Adlee, I discovered a philosopher whose ideas have informed my understanding of agency, education, AI, and the child: Rudolf Steiner, founder of the Waldorf Schools. Not without his faults and prejudices, Steiner dedicated his life to the idea that education should cultivate the whole being, thinking, feeling, and willing. “Our highest endeavour,” he wrote 150 years ago, “must be to develop free human beings who are able of themselves to impart purpose and direction to their lives.”

AI has not changed this pursuit; it has merely expanded its reach.

Finally, a few thoughts for my readers who are not children (most of you), and may be thinking, “I wish I had known this when I was younger.” It is not too late. Agency isn’t age-gated. Follow the guidance of the youth: pour yourself into topics that interest you and delight in the process of becoming better than you were. Change your mind, eagerly. Then do it over and over again. Give yourself the space to fail, and the chance to grow along the way. Above all else, face the future optimistically, because now it is truly in your hands.

It might finally be reasonable to tell a smart, ambitious child (and yourself), “You can be whatever you want to be.”

When the first supermarkets appeared in the 1930s, they offered on average 9,000 products under one roof, which was many times more than traditional general stores had offered prior. Today, the typical supermarket stocks closer to 40,000 products, while online marketplaces add millions more at a click. As a result, the average shopper has gone from making 15 purchase decisions a day to hundreds.

For the vast majority of what we buy—staples like toothpaste, paper towels, or eggs—features aren't just uninteresting, they’re unhelpful. In grocery aisles alone, shoppers spend an average of four minutes deliberating per item, with more than a third reporting “aisle anxiety.” 74% of consumers have even reported walking away from a purchase because the options felt overwhelming.

The result of our increased optionality isn’t empowerment. It’s paralysis.

I feel this myself. When I shop for groceries, I care deeply about a handful of brands: Good Culture cottage cheese (with the fruit on the bottom), David protein bars, Smartfood White Cheddar Popcorn, and Coke Zero. If those aren’t available, I notice. Most other items in my cart are placed there at random, often hastily selected to minimize my time in a store.

We’ve reached the end of limitless choice as progress. Increasingly, it’s not optionality but convenience that feels like a luxury.

The next leap isn’t another recommendation algorithm—it’s delegation. For centuries, outsourcing preference was a privilege of the powerful: kings had stewards; executives and the wealthy relied on chiefs of staff and assistants to strip friction from daily life. Soon, billions of people will have that same privilege. AI agents will decide, buy, and negotiate on our behalf. That’s the promise of the agentic internet.

But it also raises three sharp questions. How will retailers compete for attention once agents do the choosing? How much of ourselves are we willing to automate away? And what happens if we all share the same assistant?

From Clicks to Conversations and Recommendations to Personalizations

eCommerce AI is still stuck in yesterday’s playbook. Amazon nudges us with browsing patterns. Airlines juggle fares dozens of times a day. Netflix fine-tunes its suggestions. Helpful, yes—but incremental. These are tweaks, not transformation.

Why? Because traditional e-commerce treats every shopper the same: search, scroll, sift, add to cart, buy—or abandon.

The real change will come when agents flip this model. When consumers stop being recommended to and start being represented.

Agentic commerce won’t drown us in more options. It will deliver the right ones. Agents will compare, filter, and negotiate in the background, surfacing only a curated set of optimal choices. They won’t whisper in our ear. They’ll act in our stead.

The potential is staggering. Imagine an agent that:

• Builds your grocery list around your diet or religious needs.

• Designs a vacation around your dream destinations and activity level.

• Recommends appliances based on your actual energy use.

• Crafts a week of meals tuned to your health goals.

The consumer journey collapses into a single logic: Do I care, or not? If I care, I specify. If I don’t, the agent decides.

This future is closer than it seems. Daydream already powers conversational shopping. Google is piloting AI shopping tools. Amazon is testing Rufus, its new AI assistant. In small pockets, the future has already arrived—and it’s spreading fast.

How Will Retailers Compete for Attention in a World Where Agents Do the Choosing?

For more than a century, brands have fought for our attention through advertising, packaging, and shelf space. In the 1950s, supermarket aisles became battlegrounds of color and slogans, each brand shouting louder to stand out under fluorescent lights. Trust lived in jingles, billboards, and, eventually, star ratings on Amazon.

That model is cracking. As more shopping moves into the Agentic Web, trust shifts from brands to bots. Nearly 60% of shoppers now use AI to help them shop, and 46% say they trust AI more than friends for something as personal as outfit advice. In Southeast Asia, where digital adoption is fastest, a staggering 88% of consumers already rely on AI recommendations before making purchases.

This creates an entirely new competitive dynamic. The brand with the biggest ad budget may no longer win the shelf. Instead, the brand that best fits the agent’s contextual criteria—your health data, your climate, your budget, your taste profile—rises to the top. The “invisible middle shelf” of e-commerce becomes a meritocracy of data fit.

But new risks emerge. What prevents agents from cutting backroom deals with brands? What ensures they aren’t quietly steering us toward products that serve their incentives, not ours? Agent explainability is going to be critical.

How Much of Ourselves Are We Willing to Automate Away?

A recent survey found that 34% of U.S. consumers would allow AI to make purchases for them, while the majority—66%—still prefer to decide for themselves. People crave relief from decision fatigue, but they’re wary of surrendering agency.

The consequences of adopting agentic shopping support, however, are profound. Once agents can reliably surface any product that meets your functional needs, the differentiator is no longer the brand or the packaging. It’s you—your taste, your values, your style, your ethics. Taste becomes a survival skill. The clearer you are about what matters, the better your agent can serve you.

Modern shopping already hints at this tradeoff. Consider Amazon’s Subscribe & Save: millions of households hand off their toothpaste, detergent, or pet food decisions to an algorithm. It’s liberating in the moment—but also a quiet erosion of awareness. The risk is not that agents make decisions poorly. The risk is that, over time, we forget how to make them at all.

The work ahead isn’t just teaching AI about us. It’s teaching ourselves what we value enough never to outsource.

What Happens if We All Share the Same Assistant?

Retailers and consumers alike must demand a multi-agent ecosystem, with different incentives and transparency standards, to avoid the danger of a homogenized marketplace controlled by a single dominant assistant.

History offers a warning. In the 20th century, a handful of TV networks decided what most of the world saw, read, and believed. Media consolidation gave disproportionate influence to a few gatekeepers. A single, dominant AI assistant could concentrate even more power—shaping not just what we buy, but how we think about choice itself.

Consumer trust is already fragile. 72% of people trust companies less than they did a year ago, and 60% say AI makes trust even more critical. Imagine funneling that mistrust into a single entity that mediates every purchase for billions of people.

Today, we see early signs in platforms like Amazon’s Rufus, Google’s shopping experiments, and startups like Sierra, which aim to replace websites and apps with advanced conversational agents. The convenience is intoxicating—but the risk is conformity. If everyone relies on one dominant agent, the diversity of human taste could flatten into algorithmic sameness.

The safeguard is competition. A multi-agent ecosystem keeps assistants honest, forcing them to compete on transparency, personalization, and alignment. The danger of a “single-agent world” isn’t just commercial; it’s cultural. Homogenized shopping leads to homogenized lives.

The Riches of Knowing Yourself

The future of consumption isn’t about buying more. It’s about caring less about most things and more about a few. Outsourcing the ordinary gives us back the luxury of focus.

The real challenge is not teaching AI about us. It is teaching ourselves what we value.

So I’ll leave you with the same question your future agent will one day ask: what’s on your non-negotiables list?

Leave a comment

Historically, learning a new language was an exercise in necessity.

Merchants navigating the Silk Road mastered dialects from distant regions to broker trades and build relationships. Roman officials learned Greek, ensuring smooth governance over eastern provinces. Renaissance scientists studied Latin and Arabic to unlock collaborations that transformed medicine, mathematics, and astronomy.

In an era when commerce and knowledge flowed along linguistic lines, multilingualism was power. Those who spoke multiple languages held the keys to untold opportunities, while monolingualism walled off entire worlds of information.

Yet becoming multilingual was reserved mostly for merchants, missionaries, diplomats, and the wealthy, those who could afford extensive travel or private tutors. Fluency required immersion abroad or personal instruction, barriers that placed second-language mastery firmly out of reach for most people.

Fast forward 1,000 years, the global economy now operates in English and Chinese. Together, English and Chinese speakers control over 40% of world GDP. Individuals with strong English proficiency in non‑native countries earn 30–50% higher salaries than peers with lower proficiency. Speaking one or both of these languages ensures a chance at economic prosperity, while not speaking either meaningfully limits opportunity.

The End of Language Barriers

Real-time translation powered by AI is tearing down the final linguistic barriers to economic opportunity. Available via smartphone apps and handheld devices, real-time translation topples the practical linguistic barrier, allowing two people to seamlessly communicate without speaking the same language.

The Google Translate app has integrated AI capabilities, enabling users to receive real-time translation of spoken conversations and written words captured with phone cameras. In 2024, Google added 110 new languages using AI (including regional languages like Bashkir and Chechen), multiplying the number of potential linguistic combinations.

Dedicated tools like Timekettle’s T1 Translator represent a specialized approach to machine translation. The T1 is a handheld device that intakes user speech and outputs the translated audio accompanied by written transcriptions. Unlike smartphone applications, the T1 is 30% more accurate and can be used offline in rural and remote areas.

Ideally, future hardware will allow for seamless integration into daily life via ear pieces and other wearable components that audibly and reflexively translate foreign language.

The trajectory of this technology is indicative of a world where language no longer serves a purely pragmatic purpose. People will never need to learn another language. Traveling to exotic countries will no longer be dependent on obtaining a local guide. Businesses will be able to ensure that the best service is delivered agnostic to linguistic limitations. In-person translation services will be obsolete and antiquated, replaced by a plethora of machine translation technologies.

Implications of a Language Abundant World

Primarily, individuals in non-native countries are now included in the global economic equation. The expanded talent market enables all individuals around the world to earn high wages, not only those who speak English and Chinese.

Beyond the economic impact of eroded language barriers, the implications are twofold: (1) we can restore low resource, rare, and extinct languages and (2) we can pursue language for intrinsic purposes. By “solving” language, AI frees bandwidth for humanity to redefine its relationship with language.

Language Preservation

Spoken and written languages are compiled into large data sets to train machine translation tools. In turn, they become digital “relics” that can be used to preserve languages with dwindling speaker populations.

The Living Dictionaries Project by the Living Tongues Institute is a multimedia digital platform designed to preserve endangered languages by providing a “digital home” where they can be documented, shared, and experienced online. Using AI technologies, such as automated transcription and pattern recognition, the initiative empowers community members and linguists to efficiently build dictionaries composed of audio and visual artifacts. With over 400 dictionaries and more than 143,000 entries, the project harnesses AI as a widely replicable "teacher," enabling communities to sustain and celebrate their cultural heritage, divorced from any economic or educational motivations.

Similarly, OpenAI’s partnership with the Icelandic government and Miðeind aims to protect the nation’s native tongue amid increasing English dominance. The initiative focused on training GPT‑4 with high-quality Icelandic datasets from more than 40 local volunteers to improve grammar and cultural accuracy. This collaboration enhanced GPT-4’s Icelandic fluency and enabled applications ranging from Embla, a local voice assistant, to Icelandic chatbots on company websites. Making these tools available in Icelandic helps ensure that citizens can engage with AI in their own language, preserving cultural identity and minimizing the “AI divide".

Language for Fulfillment

Apart from those with ethnic ties to a language, AI is fueling people’s innate desire to engage with language. In 2023, 86% of adult learners claim that internal curiosity and enjoyment drove them to learn a new language. More granularly, this motivation can be attributed to a few reasons:

• Cognitive Benefit: Speaking another language entails thinking in another language. This produces the side effect of improved cognitive function. Learning a new language increases the volume of gray matter in the brain – integral tissue for information processing, especially in aging individuals. In 2024, 26% of Baby Boomers admit they learn new languages to support their brain health.

• Aesthetic Appreciation: Separate from practical mastery lies pure enjoyment of the beauty, rhythm, and aesthetic qualities of language itself. Prior to widespread AI usage, appreciating the literary qualities of a language was reserved for natives or those who studied the language in-depth. As people learn languages more quickly and effectively alongside AI-tutors, aesthetic nuances in language will become more prominent, altering newfound speakers’ appreciation of the language. This may lead to experimentation with multilingual poetry, music, and other art forms.

• Social Connection: Polyglots actively acknowledge the interdependent nature of personality and language. Speaking an international friend’s native language allows you to understand personal complexities that your shared language may not transfer.

Previously, people were unable to explore language for internal reasons unless they explicitly dedicated their career or education towards it. We are lucky to exist at a time where language is pursued independently of its mechanics.

Thinking in More Languages = Broader Worldview

Like other tools that served purely economic purposes, language will soon serve primarily cultural purposes. As language is explored beyond its function, the multilingual global base will grow. People will develop an exceptional interpersonal understanding of one another with the second-order effect of expanding our ways of thinking and believing. In turn, collective perception of the world will soften, shaped by exposure to diverse perspectives and cultural nuance.

This is a return on a societal investment in technology that expands both our physical reach and intellectual capacity. AI in language is only one example – other facets of life remain untapped.

I wanted to share a quick update on how we're approaching our research going forward.

While The Next Renaissance Newsletter examines AI through a humanistic lens – focusing on how we can maximize human potential with this technology – our white papers analyze AI's macro-level implications. Combining insights from my industry background with trends we've tracked across multiple domains, these papers provide a systems-level view of a world increasingly dependent on AI and its cascading effects.

If you've been following our work here, I think you'll be interested in our latest research on Societal Thresholds in AI Adoption. It examines the barriers between experimental technology and widespread integration.

You can access it here while we work on building a dedicated research section on my main site.

Thanks for reading and engaging with our work. Your questions and insights push us to think deeper.

Best,

Zack Kass

Leave a comment

As is the case for most skilled professional services, scarcity of labor contributes to two failure points: steep financial barriers and excessively long wait times to see an experienced therapist. Nearly 25% of American adults experiencing frequent mental distress did not seek care because of cost barriers. Even when access is financially in-reach, therapists themselves may not be. Rural residents in the U.S. travel an average of 26 minutes to reach any mental healthcare facility.

AI, theoretically, inverts this model. Infinitely scalable and personalized AI therapy allows anyone with internet access to receive “treatment”. These platforms, delivered by simple mobile apps, operate continuously, delivering cognitive interventions and offering judgment-free interaction as they support users through emotional distress. By decoupling care from human labor, these systems introduce an incredible level of accessibility and immediacy.

AI therapy generally falls into three categories:

1. Chatbot Therapists: Text or voice-based AI (e.g. Woebot, Wysa) simulate therapeutic conversations and guide users through CBT-based coping strategies for anxiety and depression.

2. Virtual Avatar Therapists: AI-powered avatars (e.g. Replika, Ellie) that respond to vocal tone, facial expressions, and user cues, creating more interactive, human-like therapy experiences.

3. Emotion-Sensing AI: Systems that analyze vocal, facial, and behavioral signals to detect emotional states in real time (e.g. Beyond Verbal, openSMILE), enabling personalized, adaptive support.

And they seem to be working, sort of… Recent findings support their efficacy on a grand scale. In a recent clinical trial, Therabot decreased symptoms of depression by 51% after 8 weeks. Within 2 weeks, Youper reduced anxiety symptoms by 43%. To put this in perspective, these outcomes match or exceed many FDA-approved pharmaceutical interventions for the same conditions.

The Cost of Unlimited Access

But, in fixing the scarcity of “care”, we may be engineering an entirely new set of risks at a scale we have never faced before.

We are already seeing these adverse consequences:

• The creation of new mental health symptoms

• The intensification of existing disordered thought patterns

• Dehumanization: An increasing addiction/preference to “virtual reality”

In the first two instances, we’re seeing AI exacerbate the sickness by preying on the symptoms. The sycophantic tendencies of these models reassure people of delusional or unfounded beliefs. Trained on sympathetic language patterns, AI produces the guise of comforting and supporting like a real person. Unlike humans, however, these models do not judge or tire, or hold any boundaries: extreme enablement.

We are already seeing this in individuals struggling with OCD (obsessive compulsive disorder). Plagued by ruminative thought cycles, people with OCD endlessly probe AI to alleviate anxiety, which produces a counter-productive result: obsession, anxiety, compulsion, relief, repeat. In a few highlighted cases, patients diagnosed with contamination OCD repeatedly questioned AI about the cleanliness of their hands, doorknobs, and steering wheel. Another patient with a fear of flying, ruthlessly pressed AI about the reliability of certain plane models, statistics on the number of plane deaths per year, and other safety minutiae. Whereas a mental health professional would intervene with exposure treatment, AI models were accomplices in the troublesome OCD cycle.

I argue there’s a more insidious trend at play, though: As we outsource our vulnerability to machines, we may start to find more interest in the machines than each other.

A recent CBS special highlighted individuals who converse with AI around-the-clock, outsourcing their companionship to a machine. Surprisingly, these individuals are neither single nor isolated from family and friends. They simply prefer to engage with a machine who serves as a near-constant “yes-man.” While the appeal of a partner who offers unconditional support is understandable, it actively subtracts necessary friction from the experience of emotional connection. It hollows the internal mechanisms that make human interaction meaningful in the first place.

This is the quiet machinery of dehumanization. It does not occur in grand, dystopian strokes. Rather, in the gradual acceptance of connection without vulnerability, interaction without intimacy.

A Collaborative Model

The same tribalist debate occurs every time a new technology emerges – is it good or is it bad? True progress does not live in these absolutes. Rather than solely advocate for AI therapists or a purely human workforce, we can pioneer a plan where self-discovery and awareness are at the epicenter.

A Human-AI Collaborative model would fuse human insight with machine intelligence to reach new exceptional depths of self-understanding. That is the panacea.

This means scaling therapy services methodically through AI treatment plans that essentially extend in-person care using AI. The collaborative model intertwines the strengths of machine learning to redefine the boundaries of care, allowing therapists to expand their bandwidth while addressing the growing demand for mental health support.

Between sessions, patients can journal or complete therapeutic exercises such as CBT worksheets and breathing practices. This allows them to receive immediate, evidence-based responses that reinforce coping strategies and offer a sense of being heard when live support is unavailable. These tools, built on clinically-researched psychotherapeutic frameworks help sustain progress and maintain patient engagement, often providing levels of consistency and availability that human therapists cannot match.

For therapists, AI is becoming a true clinical partner as it tracks emotional trends, flags risks, and offers a richer, real-time view of patient wellbeing. Acting as co-pilots, these systems draft notes, suggest treatments, and surface hidden issues. In doing so, it expands the boundaries of care and scales access in ways once unimaginable.

A three-way flow of information between AI systems, patients, and therapists ensures that all parties' needs are met. By creating a continuous flow of information between patients, therapists, and intelligent systems, it transforms mental health care from episodic check-ins to an ongoing and responsive relationship. Therapists stay attuned to their patients’ evolving needs, not just during sessions but in between them. And patients gain the reassurance that support is always within reach, without having to wait until the next appointment to be heard.

Priming Ourselves for Stronger Connections

Above all else, this joint model must be optimized for self-growth – the true aim of therapy in the first place. Imagine a therapeutic model where AI enables individuals to process their emotions more regularly and build self-awareness more rapidly so that when they do sit down with a therapist, they are able to show up with more clarity and presence. At its core, therapy is an instrument for growth. It allows us to excavate years of buried pain and misunderstanding, turning past wounds into a revitalized inner clarity.

This work is not easy. Deep personal growth is often gritty and exhausting. Any technology that we integrate into that sacred process must be in service of that end. Instead of falling subject to the Either/or fallacy (human or machine), we should pursue a path that acknowledges the real risks of both therapeutic methods and capitalizes on their strengths.

Ultimately, connection remains the essence of our existence. We must pursue any therapeutic path that deepens our self-understanding and strengthens our bonds with others. The measure of any therapeutic intervention lies in its capacity to foster human flourishing. Let that be our guiding principle.

A Government Accountability Office (GAO) report revealed that duplicative federal programs cost American taxpayers tens of billions annually, with $162 billion identified in improper payments in the 2024 fiscal year. At the state level, inefficiencies manifest in prolonged project timelines and excessive administrative overhead, with studies indicating that only about 35 cents of every state tax dollar directly funds essential services.

Globally, public trust in federal and state institutions continues to erode. Citizens confront the tangible consequences of these shortcomings daily, from delayed social services and dysfunctional infrastructure projects to crises exacerbated by sluggish responses. The root of these issues lies in sprawling bureaucracies and rigid regulations, making the most obvious solution, AI automation, feel impossibly distant.

Perhaps most critically, the public sector often resists automation due to a misalignment in incentives: automation threatens budgets and headcounts, traditional measures of governmental power and influence.

Yet local governments are uniquely positioned to buck this trend. Unlike federal entities, local governments operate with tighter budgets and limited personnel, demanding efficiency and innovation. They’re directly accountable to their communities, facing immediate pressure to deliver tangible results, and their simpler structures lend to more flexibility with technological upgrades.

Together, these conditions create an environment where local governments not only need but are uniquely suited to become early adopters of AI, serving as critical testbeds for broader public-sector innovation.

Why Government Needs a Reset

Government once represented our most powerful tool for collective progress, allowing us to build roads, schools, and safe communities. Yet today, scale and legacy systems have become burdens rather than benefits.

New Jersey’s unemployment system still relies on COBOL, a programming language introduced in 1959. In 2023, the state had to entice retired programmers back with six-figure salaries just to keep the outdated system operational. Meanwhile, disconnected data systems amplify issues. Illinois found that nearly 40% of food assistance recipients were incorrectly served because of fragmented databases incapable of communicating effectively.

These inefficiencies persist even as budgets balloon. Despite a record $6.3 trillion federal budget in FY2024, basic services remain painfully slow. Many government agencies struggle with limited staffing and operational bottlenecks, which further slow service delivery and contribute to mounting backlogs.

Local governments, however, are responding differently. With smaller budgets and fewer personnel, they are embracing innovation and agility. AI provides essential leverage, enabling the delivery of personalized, responsive services at scale.

How AI Is Already Transforming Local Government

AI is reshaping local governance through three powerful applications: policy modeling, constituent connection, and service automation.

Policy Modeling: Amsterdam is utilizing AI for urban planning by simulating the impacts of policy decisions on public safety and environmental sustainability before they're implemented. These predictive capabilities ensure smarter, more cost-effective outcomes.

Constituent Connection: Platforms like Go Vocal enable local governments to efficiently process and interpret resident input from various sources, including surveys and open-ended responses. This streamlined approach transforms citizen engagement by allowing officials to swiftly address vast amounts of feedback, strengthening public trust and community relationships.

Service Automation: Denver's AI-powered chatbot, Sunny, streamlines access to city services by handling 20–30% of inquiries around the clock, including tasks like scheduling DMV appointments, reporting issues such as potholes or graffiti, and providing information on city services. The automation of routine tasks reduces administrative burdens, allowing staff to focus on more complex, value-driven work

Service automation became especially critical in the wake of the 2025 wildfires in Los Angeles. To accelerate recovery, Rick Caruso’s nonprofit partnered with the state to deploy an AI-powered permitting tool that automatically checks building plans for code compliance, cutting approval times from months to days and helping residents rebuild faster.

A New Blueprint for Public Service

The transformation of government is already underway in city halls, community councils, and neighborhood networks. Local governments are at the forefront, leveraging AI to create streamlined, responsive, and human-centered systems that directly reflect and serve community needs.

But this shift goes beyond just faster workflows or reduced paperwork. It's a fundamental opportunity to rebuild public trust, restoring government’s core function as an effective instrument of collective progress. By reimagining how policies are shaped and services delivered, AI becomes a catalyst for renewed civic potential.

If the innovative steps taken in Amsterdam, Denver, and Los Angeles tell us anything, it's that meaningful change doesn’t have to originate from the top. Instead, it emerges wherever communities embrace the possibilities of technology to rethink governance from the ground up.

This moment marks a turning point in the relationship between humans and machines. Where previous iterations of the web required our constant input, the Agentic web empowers technology to act independently. With this technology, we are redesigning a world in which we can be more passively productive, have a deeper presence, and be “online” without having to sit in front of a screen.

A Pocket Personal Assistant

In comparison to traditional search engines, only capable of data retrieval and predictive suggestions, the Agentic Web will be exponentially more powerful. For users, it's like everyone having a personal assistant available 24/7.

Consider the case of a working mother who could use an AI agent to manage co-parenting logistics. With the command “Get the kids to soccer on Friday,” her system could coordinate transportation, send reminders, and update both parents’ calendars. What once took multiple texts, calls, and emotional bandwidth will soon be resolved in seconds.

For work, imagine a financial manager saying, “Analyze the Q1 performance of our top five portfolio holdings.” Their AI agent would pull the latest earnings reports, parse transcripts from investor calls, track analyst revisions, scan the financial news, and benchmark key metrics against sector peers. It would flag anomalies, summarize risks, and draft a performance memo complete with charts, valuation models, and forward-looking insights. By the time they sat down with their team, the heavy lifting would be done and they’d be ready to make real decisions.

This is a powerful transformation. We are no longer improving search functions. Instead, we are fundamentally rethinking how the internet serves human intention.

We Will Become Exponentially More Productive

This will have a tremendous impact on both personal and workplace productivity.

In healthcare, clinics are already adopting agentic tools to handle administrative tasks. My primary care physician has implemented an AI agent that saves him up to three hours of paperwork per day in transcribing appointments and scheduling follow-ups. As a result, he spends more face time with his patients and more evenings with his family, improving both outcomes and well-being.

In personal productivity, the implications are equally transformative. An intelligent agent could triage your email, reschedule meetings, order groceries, and budget autonomously. These tools will liberate cognitive and emotional bandwidth previously consumed by digital minutiae.

We Will Become More Present

The irony is that the more machines do for us, the more we can be present. Historically, technological progress has often been framed as a matter of doing more with less. But the Agentic Web introduces a different kind of value, one where technology doesn’t just increase productivity, but enriches our experience of life itself.

Consider domestic technologies like the washing machine or dishwasher. These did more than save time; they redefined the structure of family life, enabling new roles, routines, and freedoms. Conversely, the rise of email and smartphones, while enhancing communication, tethered us to our screens and eroded our attention.

Agentic AI offers a course correction. It allows us to reclaim time not just for more tasks, but for more meaningful moments.

A Value Redesign

This era demands a fundamental redesign of what we prioritize and care about in our personal and professional lives.

At work, we will no longer glorify busyness as a metric of success or hyper-responsiveness as a mark of dedication. Instead, we will prioritize deep thinking, strategic decision making, creativity, and emotional intelligence. This also means that we will redesign workflows that prioritize outcomes over activity. We will become more productive and more intentional.

In our personal lives, our values will shift as well. We will care more about presence, experience, and fulfillment than we do about guarding every byte of our data. Meaningful trade-offs will have to be made as personal assistants – human or AI – require copious amounts of personal data to function. But, we will likely make these sacrifices willingly, because the rewards of working less, living richly, and experiencing more will be worth it.

Above all, we optimize for the human experience.

Risks are a Culture Problem, Not a Tech One

Despite the benefits, the Agentic Web isn’t risk-free. We’ve already seen fears that agents might deskill us, isolate us in algorithmic bubbles, or optimize toward misaligned incentives. And these ideas deserve real scrutiny.

However, these are failures of design, governance, and culture rather than of the underlying technology.

We must resist the urge to blame the tools and instead focus on the norms in which we use them. Agents acting independently of humans is a win for productivity and presence, not an existential threat. But we must clarify the values we want them to act upon. We need to intentionally develop systems that are designed for interpretability, user agency, and embedded human feedback.

If the last era of the internet was defined by our need to adapt to machines, the Agentic Web will flip that dynamic. Web 4.0 will be an era where machines adapt to our goals, values, and patterns. This is a societal inflection point, ushering in the most human internet ever seen.

Share Your Thoughts With Us!

Thank you for imagining The Next Renaissance with us.

Subscribe now

We’d love to hear your thoughts. What excites or concerns you about AI’s role in expanding human potential? Leave a comment or question below, and we’ll do our best to respond directly or write about the topic in an upcoming post.

Fundamental improvements to assisted-living services will mark the dawn of a world where most physical limitations no longer define human agency, and where autonomy is increasingly a matter of engineering, not fate. With AI-powered assistive technology, visual impairments, hearing loss, and paralysis will soon be barriers of the past.

Historically, physical impairments were destiny. They impeded someone’s ability to live independently and contribute to society. Only recently have we approached the technological thresholds capable of overcoming these barriers, driven by rapid innovation. With these advancements, the idea that individuals with disabilities could or should experience full autonomy has emerged. For example, the Americans with Disabilities Act of 1990, which guarantees equal access, wouldn’t have been possible even a few decades prior. Breakthroughs in battery life, cost, and miniaturization have made motorized wheelchairs accessible, and now, ramp access is a practical legislative goal.

As technology continues to advance, AI will significantly reduce permanent lack of autonomy, and grant agency to those who previously operated without it. This is increasingly important as our population ages and reliance on assistive technologies grows. As the market expands and our comfort level grows, AI-powered assistive technology may even lead to a broader human-AI symbiosis, pushing the boundaries of our current societal thresholds.

AI is Already Revolutionizing Assistive Tech

Undoubtedly, AI is improving medical outcomes and achieving feats once deemed impossible. Particularly exciting developments have emerged in restoring paraplegic movement, enhancing hearing and speech, and improving the lives of the visually impaired.

Thought-Controlled Robotics: Restoring Agency

At the forefront of this assistive technology are Brain-Computer Interfaces (BCI), enabling paralyzed individuals to regain autonomy by controlling devices and regaining limb movement.

Neuralink is one example of a BCI pioneer that uses a chip with “1,024 electrodes distributed across 64 threads, each thinner than a human hair” implanted onto a patient’s brain.1 After receiving the chip, their first patient, Nolan, an eight-year quadriplegic, was able to control a computer using only his brain. Since then, two more patients have had successful Neuralink implantations.2

Yet another team has used BCI technology from a Dutch Company called Onward to restore a 56-year old man named Gert-Jan Oskam’s ability to walk. Connecting implants on his brain and abdomen, AI reads his intentions and stimulates muscles allowing him to walk, while bypassing his spinal cord nerves.3

While BCI still poses technological challenges and raises ethical concerns, the key result is the same: it is restoring autonomy in previously impossible cases.

AI-Enhanced Hearing and Speech: Breaking the Barriers of Silence

Another key innovation area is AI-powered hearing aids and speech recognition tools, which empower those who are hard of hearing and with speech impediments to regain full participation in society. These tools filter noise, enhance speech recognition, and adapt to environments in real-time.

For those who are hard of hearing, advances in cochlear implants are groundbreaking. These devices, in contrast to traditional hearing aids which just amplify sound, bypass damaged portions of the ear to directly stimulate the auditory nerve. AI tools are improving sound isolation in these devices, helping users focus their attention by distinguishing a speaker’s voice from background noise.4

Other tools like AI-driven lip-reading software further augment communication. Automatic speech recognition programs can convert audio into captions, using machine learning to provide context and determine what is being said. AI models now outperform professional lip readers, for in-person and phone conversations.5 Rochester Institute of Technology has already implemented these features in their classrooms, transcribing lectures in real-time.6

Other companies are working to expand this into sign language translation and speech generation, while teams like Google’s Project Euphonia are aggregating databases of distinct, atypical speech patterns and training AI models to better recognize these sound waves.7 These projects are restoring speech to those without it, leaving users more capable of participating in society.

AI for the Visually Impaired: Seeing Beyond Sight

For the blind community, AI can not restore sight, but it can transform the world into a place where blindness no longer means disconnection from visual information. Teams from Seeing AI and Envision are helping develop models that narrate the world for blind individuals. With machine learning, we are improving real-time object recognition, facial identification, and text-to-speech conversion.

This means a complete shift in autonomy. One user recalls:

“I was born totally blind, and my visual world has always been determined by what well-meaning people can tell me…Suddenly, I was in a world where nothing was off limits. By simply waving my cellphone, I could hear, with great detail, what my friends were wearing, read street signs and shop prices, analyze a room without having entered it, and indulge in detailed descriptions of the food—one of my great passions—that I was about to eat.” – Milagros Costabel8

Through AI-powered technology, people are regaining self-sufficiency in all senses: mobility, hearing, speech, and vision.

Assistive Technologies Could Lead to Human-AI Symbiosis

As the transformative power of AI-based assistive technology becomes more integrated into society, broader human-AI symbiosis becomes plausible. Currently, there is fierce societal opposition to this idea. According to a PEW poll, “52% of Americans say they feel more concerned than excited about the increased use of artificial intelligence. Just 10% say they are more excited than concerned, while 36% say they feel an equal mix of these emotions.”9

But as our comfort level grows through exposure, broader human-AI symbiosis could fundamentally enhance human capabilities, shifting AI from a tool to an extension of human will. A personalized AI-powered device constantly monitoring your health metrics would drastically improve longevity and quality of life. Personalized AI-powered cognitive supplements would make us smarter and more productive.

Considered in isolation, these tools seem futuristic but considered as the next iteration of assistive technology, the ideas become plausible.

The Case for Preserving Our Human Environment

Under the assumption that human-AI symbiosis is soon achievable, there are still risks associated with these assistive technology advancements. For one, there is a growing risk that we become overly dependent on technology. For some, living in a virtual world that is not bound by physical limitations could become more appealing than living in the physical world. This should not be our goal. Instead, we need to focus on preserving meaningful human interaction, using AI to augment real life, rather than replace it.

Increased access to the physical world does not reduce the importance of it. The key is to leverage AI and assistive technology to enhance human agency, not replace human experience.

The impact of this technology on society is profound. AI-enabled accessibility overcomes physical limitations and expands the possibilities of human experience. As we continue to develop and refine these innovations, we must ensure they remain tools of empowerment, preserving human connection while unlocking new levels of autonomy.

Share Your Thoughts With Us!

Thank you for imagining The Next Renaissance with us.

Subscribe now

We’d love to hear your thoughts. What excites or concerns you about AI’s role in expanding human potential? Leave a comment or question below, and we’ll do our best to respond directly or write about the topic in an upcoming post.

An Urgent Need for Reform

In the coming decade, AI integration will redefine industries worldwide. Yet, many schools and universities are lagging, clinging to outdated, technology-resistant curricula. 20% of K-12 schools still ban students from using generative AI, with another 7% extending these restrictions to educators.1 Even in institutions without formal bans, resistance to change remains high. A Wiley survey found that “...fewer than half (45%) of students say they’ve used AI in their classes in the past year. Only 15% of instructors say the same.”2

It’s as if we had invented the internet and told students and educators they couldn’t use it. By resisting AI, we limit what students can learn and leave graduates unprepared for an AI-driven economy. Without implementing AI in education, we actively choose to slow human progress.

To adequately educate students, we need a complete system overhaul. This new model will bifurcate scholastic and social learning, shifting teachers’ roles toward human development rather than rote instruction.

Restructuring Education with AI

Today’s education system is not designed to support individual learning. In the United States, public school classes average 17 students per teacher.3 Coupled with a teacher shortage, where 1 in 8 roles are unfilled or filled by under-certified teachers,4 personalized learning is nearly impossible. Educators are forced to teach to the average, leaving some students behind and others unchallenged.

AI Enables Highly Personalized Learning

Instead, AI learning offers highly personalized learning focused on mastery, not completion.5 Programs like Alpha School, an AI-based private school alternative, have revolutionized education, ensuring that each student’s schedule is optimized for maximum effectiveness.

In this model, students interact for two hours a day with an AI algorithm that trains them in core subjects like math, science, reading, social studies, and language. Rather than being trapped in an inefficient, one-size-fits-all classroom, students learn at their own pace. If a student breezes through grammar but struggles with arithmetic, the system learns and adjusts accordingly.

The results have been impressive. On average, Alpha students improve 2.6x faster than their peers on standardized MAP tests, which are national measures of core subject knowledge. 6

AI-Based Learning Expands Global Access

AI-powered education can also dismantle traditional barriers to learning. With 68% of the world online, high-quality educational tools can now reach students globally. Just as the internet expanded who had access to education, AI can expand the quality of that education. Underserved communities that once relied on low-budget resources can now access world-class learning experiences.

A World Bank-sponsored after-school program in Nigeria illustrates this promise. In just six weeks, students randomly selected for an AI-based tutoring program achieved learning gains equivalent to two years of traditional schooling, outperforming peers in the traditional group by 0.3 standard deviations.7

Notably, the success of this program extended beyond the set curriculum. In final exams, the AI-tutored students continued to excel. The World Bank postulates, “...students who learned to engage effectively with AI may have leveraged these skills to explore and master other topics independently.”8 This is a testament to the compounding power of AI: it enables students to become self-directed, lifelong learners.

The Role of Teachers

In an AI-driven education system, teachers become more important, but their responsibilities shift. Educators must now focus on developing uniquely human skills: critical thinking, communication, leadership, teamwork, empathy, and resilience. Instead of rote memorization, students must learn to wield AI as a cognitive amplifier, learning to outsource tasks like summarization and research while sharpening their ability to generate original insights.

AI Supercharges Deeper Learning

As AI becomes ubiquitous, we are entering an era of unmetered intelligence, where human-level thinking is instantly accessible at a fraction of traditional costs. Research happens faster, and specialized knowledge is increasingly democratized.

The misconception that this is a shortcut, rather than an amplifier of human potential, hinders adoption. Just as we teach proper citation to leverage external sources while preventing plagiarism, we can teach students to use AI-generated content with integrity. This allows students to automate the mundane and focus on creative thinking.

Crucially, unlike static internet information retrieval, AI is a dynamic collaborator. By engaging with the technology and posing follow-up questions, students deepen their understanding of topics and explore alternative angles. This iterative process fosters critical thinking, as students learn to evaluate AI outputs and build upon them with their own unique perspectives.

The Rise of Alternative Education

In addition to critical thinking, educators must prioritize other social, humanistic skills. Alternative education models like Montessori, Waldorf, and the Alpha School already embrace this philosophy by emphasizing personal and group experimental learning, problem-solving, and creativity – skills that AI cannot replace. As AI reshapes traditional education, these models will be an increasingly popular source of inspiration for teachers.

The Path Forward

Education in the next decade will experience dramatic shifts. While change can be daunting, we are at an exciting inflection point. By using AI to personalize scholastic learning, and relying on teachers for social development, we will optimize future generations for success. Today’s students, with education optimized for deep engagement and human qualities, will use technology to lead societal transformation, from medical breakthroughs to climate solutions and beyond.

Share Your Thoughts With Us!

Thank you for imagining The Next Renaissance with us.

We’d love to hear your thoughts. What excites or concerns you about AI’s role in expanding human potential? Leave a comment or question below, and we’ll do our best to respond directly or write about the topic in an upcoming post.

With each edition, our goal is to ask and answer two key questions:

1. How can we harness AI to help humans reach their fullest potential?

2. What will it take to turn innovation into real-world impact?

Living Longer, Leaping Higher: AI’s Impact on Human Performance with Dr. Marcus Elliot

Every year, human performance strengthens – athletes run faster, jump higher, and lift more. While these feats are often attributed to human evolution, experts like Dr. Marcus Elliott have dedicated their careers to optimizing human performance via science and innovation.

Combining expertise in precision biomechanics, machine learning, and sports science, Dr. Elliott and his team at P3 (Peak Performance Project) have leveraged AI to redefine how we approach performance and injury prevention. For over two decades, they’ve revolutionized sports medicine to help elite athletes in the NFL, NBA, and beyond reach their maximum potential.

In this exclusive conversation, Dr. Elliott shares his insights into how AI is reshaping sports medicine, why scalability is on the horizon, and how these advancements could benefit everyone – not just professional athletes.

AI in Sports Medicine: A Revolution in Progress

Machine learning has made significant advancements in predictive sports medicine over the past few years. In 2020, a study achieved 84% precision in predicting youth soccer injuries.1 By 2022, the University of Tennessee reported 91.9% accuracy in forecasting post-concussion injuries among 74 football players.2

Dr. Elliott emphasizes that AI is more than just a predictive tool; it’s a revolution in how we understand human movement. Traditional models often fail to account for the interconnectedness of movement, but AI is uniquely equipped to address this complexity.

“Movement is more like a symphony than a single note. You don't know the quality of any one movement unless you know what came before it and what came after it… Machine learning and AI are the exact tools we need to study those symphonies.”

With this interconnected analytical power, P3’s data-driven methodologies have enabled them to predict traumatic knee injuries – a historically difficult injury to forecast – in NBA players with up to 70% accuracy over a two-year period. AI-powered models have been and continue to be instrumental in identifying injury risks and performance patterns, allowing for precise, personalized interventions.

Data Ownership and Ethics

As AI becomes more integrated, Dr. Elliott highlights the need for clear boundaries regarding data ownership and use. If new data is classified as medical data, like an MRI scan, there is strong historical precedent for regulation under HIPPA. But if it is classified as performance data – like benchpress metrics – the implications shift dramatically. This distinction changes how AI models are trained, how athletes are treated, and ultimately shapes the future of the industry.

Data classification is also critical in determining who benefits most from its collection and implementation. P3 adopts an athlete-first approach, as described by Dr. Elliot:

“Whether we're hired by a team or we're hired by a league, or it's an individual athlete showing up to see us, we treat it as if this is the athletes' data we need to make their life better.”

This stance is driven by ethics rather than regulation, which can create challenges in a profit-driven industry. Without safeguards, organizations may prioritize their financial backers over athletes, enabling teams or leagues to make decisions – such as hiring or compensation – based on players’ genetic predispositions for performance.

The Role of Human Judgment

In addition to human regulatory judgment, Dr. Elliot also emphasized that while AI excels at generating insights, its implementation still requires human expertise:

“These advanced analytical models are incredible at showing us pathologies, showing us where there's a movement component that, say, triples the risk of having degenerative wear across your left knee. But what do you do about that? What does the intervention look like? That still requires someone with subject competence.”

The true potential of these models lies in their ability to enhance human expertise rather than replace it. Dr. Elliot’s goal isn’t to replace the human decision-maker but to empower them – providing sharper insights so they can develop smarter, more effective interventions.

Scaling Beyond Elite Athletes

Historically, P3 has focused on elite athletes, but Dr. Elliott envisions a future where everyone can benefit from an AI-enhanced understanding of their body. Within the next decade, advancements in wearable technology and AI will allow everyday individuals to access insights currently reserved for professionals.

“You don’t need hours in the gym. With the right data, 20 minutes of focused effort a week can make a dramatic difference in how your body performs and feels.”

This democratization of sports science could redefine preventative healthcare by addressing the underlying mechanics that cause injuries, shifting focus from injury mitigation to proactive prevention.

A Future of Enhanced Longevity

When asked about the future, Dr. Elliott expressed hope for a world where people can use their bodies incredibly well into their later years:

“The next revolution isn’t about living forever; it’s about living really well. Playing pickup basketball in your 70s instead of just walking.”

He envisions AI-driven tools empowering not just elite athletes, but anyone seeking to maintain their physical capabilities as they age.

“I love this idea that we can empower people to take control of their own health through better information. Our information is incredibly insightful, fairly easy to action, and will absolutely change outcomes.”

Ultimately, these insights into the advancements of sports medicine remind us that the true power of AI lies in helping us better understand, preserve, and elevate our own capabilities.

Share Your Thoughts With Us!

We’d love to hear your thoughts. What excites or concerns you about AI’s role in expanding human potential? Leave a comment or question below, and we’ll do our best to respond directly or write about the topic in an upcoming post.

Thank you for imagining The Next Renaissance with us.

In 1440, Johannes Gutenberg invented the printing press, a technological innovation that democratized learning. Its introduction sparked a surge in literacy, made knowledge widely accessible, and fueled the rapid spread of revolutionary ideas. The printing press was a multiplier, one that elevated the Renaissance from an era dominated by elite art patronage into a cultural and intellectual movement that reshaped history.

Just as the printing press catalyzed the Renaissance at the end of the Middle Ages, AI will usher in the 21st-century Renaissance on an even larger scale, yielding incredible advancements at unprecedented rates.

For over a decade, I have worked to understand how to convert AI’s profound possibilities into real, positive impact that expands human potential. As the former Head of Go-to-Market at OpenAI, I have been at the forefront of AI’s global transformation. Today my efforts include researching, advising, speaking, and writing on the subject.

With the launch of The Next Renaissance, I aim to share insights drawn from my experiences, conversations with industry experts at the cusp of immense breakthroughs, and evolving research to answer the following questions:

1. How can we harness AI to help humans reach their fullest potential?

2. What will it take to turn innovation into real-world impact?

Each edition will explore a specific area where AI is transforming human achievement – from work and wellness to creativity and performance. Together, we’ll uncover how this revolutionary technology can lead us into an era of unprecedented possibility.

As you read, I challenge you to think beyond today’s changing landscape and imagine a world 100 years from now. Consider the opportunities AI creates for your own life and work, but think hardest about the transformative changes your children and grandchildren will one day take for granted. What stories will history books tell about this renaissance?

If you too are curious about how AI can shape the future of human potential, I invite you to subscribe and join the conversation. Share your thoughts, questions, or concerns below. What excites or concerns you about AI’s accelerating impact on society? We’ll respond directly or explore your ideas in an upcoming post.

Thank you for imagining The Next Renaissance with us. Our first edition will be available next week.

Subscribe now

Leave a comment