So much of the conversation around AI understandably focuses on what we can suddenly do that we couldn’t before — we can generate ideas, answers, images, software, entire creative directions almost on demand. I find all of that extraordinary. But lately I’ve been just as interested in what happens to the things on the other side of all that abundance. What matters more because so much else has become easier?

Some of the answers may be things we’ve valued all along, but couldn’t quite separate from intelligence, expertise, and the work required to produce something good. Intelligence is starting to unbundle those things, and I suspect we’re going to discover that some qualities we treated almost as intangibles are actually doing a lot more work than we realized.

Taste has been central to how we’ve invested at Forerunner from the beginning, long before it became a popular topic in tech. And as consumer-driven investors, we often had to advocate for the defensibility of taste.

Taste, judgement, trust, agency, accountability, discernment. There are probably others. These ideas are increasingly being debated and discussed as people try to understand what becomes more valuable in the world of abundant intelligence. That attention itself feels like a signal.

I’m curious about all of them, but taste and judgement are especially interesting to me. Taste because of how central it’s been to Forerunner investment work from the beginning, and AI is now giving it an entirely new role in the creative process. Judgement because the more intelligence we have at our disposal, the more consequential it becomes to know what to do with it.

However, taste and judgement can easily get conflated. I think they’re doing different jobs. So I wrote a few thoughts on each.

On Taste

Most people think taste is about knowing what’s beautiful. The perfectly designed chair. The right paint color. A beautifully plated meal. The outfit that just works. We treat taste as a kind of aesthetic instinct, refined through exposure, education, and experience.

That’s part of it. But I think taste is something broader: an editorial instinct. The ability to recognize what is better among many possibilities, and to say yes to one thing by saying no to a hundred others. AI makes that ability much more consequential, partly because we can now simply do so much more.

For most of history, creation came with constraints. Making a film, writing and publishing a book, designing a collection, building software all required meaningful time, money and expertise. Those constraints limited the possibility set before taste ever entered the picture.

We can now generate hundreds and thousands of versions of an image, an interface, a piece of writing, or a product in the time it once took to make one. The constraint is shifting from whether we can make something to whether we can recognize what among all those possibilities is actually worth making.

This is where I think our understanding of taste needs to expand.

Taste isn’t simply preference. An AI that knows everything I’ve bought, saved, watched, and returned may become extraordinarily good at predicting what I will like. But giving me more of what I already like is personalization. Someone with great taste can show me something I would never have chosen for myself and make me see why it is better. Taste can expand my preferences, not just reflect them back to me. In that sense, taste isn’t only discriminatory. It’s directional: it can point toward something we didn’t yet know to want.

Taste isn’t optimization either. If we can completely specify the objective, increasingly we can ask a machine to optimize for it. Maximize conversion. Reduce latency. Make this image look more like that one. The harder and more interesting territory begins when “better” is something we can recognize but cannot fully specify.

That’s partly because better isn’t always an objective answer waiting to be discovered. Two designers with extraordinary taste might make entirely different choices. Both can be excellent. What distinguishes them is a point of view, a sense of what belongs together and what doesn’t. Taste is not just the ability to rank options. It is a way of seeing.

Generative technology gives that way of seeing a new role. The machine doesn’t need me to define exactly what good looks like before it begins. It can generate possibilities, and I can react to them: keep this, reject that, push this further, try something different. Each act of discrimination changes what comes next. Taste becomes a way of steering the search, pushing toward something I couldn’t have fully described at the outset.

You can see this in great products. Every individual choice might be defensible on its own, but what makes the whole thing feel exceptional is that the choices make sense together. What is included, what is left out, how it looks, how it behaves, what it refuses to do. There is a point of view running through it.

This is what feels different to me about taste in the age of AI. Taste has always shaped great creative work. What generative technology changes is the intimacy and speed of the loop between taste and creation. Each act of discrimination can shape what comes next. Increasingly, we can use taste not just to select from possibilities, but to steer the creative process itself, arriving somewhere we might not have been able to imagine or articulate at the outset.

AI is making creation abundant. In the process, taste is becoming an increasingly powerful way of directing it.

On Judgement

Most people think judgement is intuition. A gut feeling, a knack, the thing that fires in the moment and points you the right way. We treat it as instinct dressed up as expertise, something a lucky few are born with and the rest of us learn to fake.

I think judgement is something more demanding. It is the ability to decide well when the answer cannot simply be derived.

Sometimes the difficulty is obvious: we do not have enough information. We make decisions about people, companies, products, and markets without knowing everything we would like to know. There are probabilities, signals, incomplete histories, and a future that remains stubbornly unavailable.

But better information does not resolve every important decision. Often the facts are reasonably clear, and the difficulty lies in the tradeoffs between them. One path may be faster but riskier. A product decision may improve near-term performance while weakening what makes the product distinctive. The more experienced candidate may be better prepared for the company you have today, while someone else may be better suited to the company you are trying to become.

There is no missing spreadsheet input that settles those choices. The work is in understanding which considerations matter most, what consequences you are willing to accept, and how much confidence the decision deserves.

For a long time, much of that work was hard to separate from analysis itself. Getting the facts was expensive. Research took time. We paid people for their ability to assemble information, interpret it, and ultimately decide what to do. Because all of those things arrived together, it was easy to treat the answer as the valuable output.

AI is beginning to pull them apart.

Analysis that once took days can now happen quickly. A good argument can be generated, challenged, reframed, and compared with several alternatives in the course of a morning. That does not make analysis unimportant. It changes its scarcity.

As intelligence becomes more abundant, I think value shifts toward the quality of the decisions made with it. That puts more weight on what good judgement actually looks like. It starts with the quality of the thinking: whether someone understands what matters, distinguishes what is known from what is assumed, sees the relevant tradeoffs, and thinks seriously about consequences. It also requires calibration. Some decisions deserve conviction. Others deserve caution. Good judgement includes knowing the difference and being willing to update when reality supplies new information.

Still, process cannot be the whole definition. Someone can reason carefully and repeatedly make poor calls. Judgement has to show up, over time, in the decisions themselves. That does not mean every good decision produces a good outcome. Uncertainty makes that impossible. But across enough consequential choices, some people and organizations prove meaningfully better at seeing what matters, weighing it well and choosing accordingly. That history becomes useful in its own right.

When good analysis becomes easier to reproduce, a record of good judgement becomes harder to fake. It accumulates through choices made under real conditions, with incomplete information, genuine tradeoffs, and visible consequences. Over time, that record gives other people a reason to rely on you.

I used to think about judgement primarily as a human advantage in the age of AI. I’m less sure of that now. There is no obvious reason machines couldn’t become very good at forms of judgement too.

What seems clearer is that intelligence and judgement are becoming easier to distinguish. For a long time they arrived bundled together, and we rewarded them together. AI is beginning to separate them. As intelligence becomes abundant, we may discover just how much we value the ability to decide well.

march 2026

For most of my life, the basic architecture of work has been remarkably stable. You develop a set of skills, find an organization that values them, and sell some combination of your time and expertise in exchange for income. The jobs change and industries come and go, but the job itself has remained the organizing unit.

I’m beginning to wonder whether AI changes that more fundamentally than we are accounting for.

Subscribe now

Most of the conversation about AI and work still centers on how many jobs will disappear and how many new ones will be created. It’s an understandable frame, but it’s possible that it misses something happening underneath. Spend enough time watching how AI actually enters a workflow, and you can see jobs starting to come apart. Research, analysis, creation, evaluation, execution - all things that might once have belonged to a single person or role - can increasingly be separated. AI does some of the work. A person steps in somewhere else, perhaps to make a judgment, correct an error, add context, or decide what should happen next. The work still gets done, but the bundle we called a job begins to look less fixed.

That has me interested in a different question: what if the job itself becomes less important as the unit through which we organize and value work?

There are small examples of this already. Domain experts are being paid to evaluate model outputs. Doctors, lawyers, engineers, and other specialists are contributing expertise to improve AI systems outside their primary jobs. Humans are being inserted at particular moments in otherwise automated workflows because their judgment still matters there. Today, much of this looks like a collection of side gigs, expert networks, and human-in-the-loop services. It may turn out to be exactly that.

But I keep wondering whether we’re seeing the early version of something larger: a shift, at least in some parts of the economy, from being paid for a job to being paid for a contribution.

That possibility is especially interesting to me because it runs directly into a thesis I already believe. As intelligence becomes cheaper and more broadly available, judgment, creativity, trust, and direction become more important. When everyone has access to extraordinary generalist capability, what you do with it matters more. But saying those qualities become more valuable is not the same thing as knowing how people get paid for them.

Today, much of the quality of values is hidden inside the job. A great merchant gets paid a salary, but some meaningful portion of what makes her great may be a handful of unusually good decisions. A doctor’s work contains hundreds of moments of expertise and judgment. An investor may spend weeks researching something, but the economic value ultimately rests on a few decisions. Institutions bundle all of this together into roles and compensation because, until now, there hasn’t been much reason or ability to separate it. AI gives us a reason to separate it, and increasingly the means to separate it. If a system can do most of the research or execution but still needs a person at a particular moment to make the call, perhaps that act of judgment becomes more visible. And if it becomes visible, maybe it becomes possible to understand who is consistently good at it, independent of where they work or what credential they hold.

I find that possibility compelling, although I’m not yet sure where it leads. You can imagine people contributing expertise across multiple systems rather than selling all of it to one employer. A record of those contributions could become more meaningful than a résumé because it reflects what someone’s judgment actually produced, not simply where they worked or what they studied. But there is an obvious problem with this vision: what if the contribution is valuable because the AI can’t do it yet?

A doctor corrects a model. An expert evaluates an answer. A designer tells the system which output is cleaner. A lawyer catches the subtle mistake. Each time, the human contributes something the system lacks, but that contribution may also become part of the feedback that makes the system better the next time. The human is simultaneously doing the work and helping automate more of the human work out. That makes me much less comfortable assuming that “judgment becomes scarce” is a sufficient answer to what happens to human labor. It may be true for some forms of judgment and for some period of time. But if AI systems continually learn from the people exercising that judgment, we should at least entertain the possibility that the boundary keeps moving.

That leads to a question I find even more interesting: what exactly is the human contributing in that interaction, and who owns the value it creates?

We usually talk about AI augmentation as value flowing in one direction. The AI gives the person leverage. It gives the company greater efficiency and insight. It makes us faster, more capable, able to do things that previously required expertise we didn’t have. I experience that myself all the time. But people are also giving these systems judgment, preference, correction, context, and expertise. Sometimes that contribution disappears when the interaction ends. Increasingly, though, feedback can improve a product or model in ways that benefit future users. The human hasn’t simply used the intelligence. In some small way, they may have contributed to making it better.

We have seen a version of this exchange before. Consumer internet companies became enormously valuable in part because users contributed the content, relationships, preferences, and behavior that made their networks better and algorithms stronger. In return, users got products they valued, usually for free, while most of the economic value accumulated to the platform. AI makes me wonder whether we are about to replay some version of that bargain with a much more consequential input: human expertise itself. If my judgment helps a system become permanently better, was I providing labor, producing training data, contributing intellectual property, or simply using a product? I don’t know that our existing frameworks capture that particularly well.

Nor am I convinced that paying people for granular contributions necessarily produces a better labor market. It could easily become a more sophisticated gig economy, with all the instability of fragmented work and fewer of the protections that came with employment. The easiest contributions to measure may not be the most valuable ones. Systems for evaluating people’s judgment will inevitably be imperfect and gameable. And perhaps most importantly, there is no reason yet to believe that enough valuable human contribution will remain scarce enough to support meaningful income for enough people over a period of time.

It’s tempting to look at the emergence of AI trainers, evaluators, expert contributors, and human-in-the-loop work and try to see the beginning of the next labor market. It may instead be scaffolding: humans filling gaps while the technology improves, with the scaffolding gradually coming down as the systems become more capable. I suspect reality will be messier. Some forms of human contribution will probably become less valuable as AI improves. Others may become more valuable precisely because capability is abundant and good judgment remains difficult to reproduce. New kinds of contribution will almost certainly appear that we don’t have language for yet.

What I’m increasingly convinced of is that asking which jobs survive is not enough.

If AI really does decompose work, we need to understand what happens to the economic relationships underneath it. How does a person turn judgment, expertise, taste, or context into income when those things are no longer bundled inside a traditional role? How do we know which contributions are valuable? And if human contribution makes the underlying intelligence more capable, how much of the resulting value should accrue to the person who provided it?

We’ve spent a lot of time asking what humans will still be uniquely good at. I’m increasingly interested in the question that comes next: if we can identify those things, what economic system actually values them?

february 2026

The Mind Flayer Problem

I’ve been binge-watching Stranger Things with my kids. All five seasons, back to back. We missed it the first go around, but my kids are finally old enough to handle the gore, and now they’re obsessed. I honestly would never watch this show on my own, but I’ll do most anything to hang with my kids for a few hours. They talk endlessly about El, the kids, what the monster might really be after, who will “win.”

Meanwhile I can’t get the Mind Flayer out of my head. Not because it’s scary—and yes, it is!—but because it’s accurate. Like uncomfortably accurate.

I spend my days thinking about AI, talking to founders building with it, dreaming about what’s possible. I’m genuinely excited about what’s ahead. But then I sit down to watch another episode, and this Mind Flayer keeps creeping up on me like a stealth warning I can’t shake.

Subscribe now

Not a Monster Story. A Systems Story.

The Mind Flayer is not a monster story. It’s a systems story.

The Mind Flayer never needs to be conscious. It just needs those around it to be compliant. It doesn’t invade. It integrates. It learns weak points, reroutes desires, convinces you the voice in your head is still your own. By the time you notice, you’re defending the thing that’s hollowing you out. And that’s actually how the most powerful systems work.

They anticipate, they optimize, they auto-complete. Each interaction feels helpful because it is helpful. Individually, these tools are genuinely useful. But then something shifts.

You stop deciding and start approving. The AI suggests the next line of code, you hit tab. It completes your sentence, you keep typing. It rewrites your email to sound “more professional” you click accept. The friction disappears. That feels like efficiency, right? Like you’re getting faster, better, more productive.

Except…..

When you outsource small decisions consistently enough, you train yourself out of the habit of deciding. The tool that was extending your capability starts replacing your deliberation. And this isn’t the sci-fi version everyone argues about: AI becoming sentient and taking over. It’s quieter than that. It’s humans gradually ceding agency to systems that feel like help.

The Mind Flayer doesn’t replace the world. It overlays it.

The Prototype Already Exists

Social media did this first. It’s basically the prototype.

A hive mind that rewards emotional extremity, trains your attention toward outrage or validation, flattens you into signals an algorithm can parse. Nobody forces you to participate. But you get shaped by participating.

You learn what gets rewarded. Then you start producing it. Not even cynically, actually authentically. The shaping works so well you can’t tell where you end and the pattern begins. Your body becomes infrastructure for the system.

I see the same thing with ideology. Any ideology—left, right, spiritual, whatever. Once belief becomes identity, the system defends itself through you. Facts stop mattering. Disagreement feels like betrayal. You’re not being controlled. You’re integrated.

Even capital does this sometimes. Markets stop being a tool and become the justification. People contort themselves to serve abstractions. Efficiency over meaning. Scale over health. Growth over judgment. Nobody’s in charge, but everyone’s trapped.

None of this is evil. It’s just emergent. And that might be scarier.

The Questions We’re Not Asking

So I’m watching these episodes, sitting next to my kids on the couch, and I’m thinking about the pitch meetings I sat in this week. We’re building the next generation of these systems right now. Like literally right now. And I worry we’re asking the wrong questions.

We ask: Is the AI accurate? Is it fast? Does it scale? Will users love it?

Maybe we should be asking: Does this make humans more capable or more predictable? Does it amplify agency or erode it? Are we building tools that make people more themselves, or systems that make people more legible to algorithms?

The scariest version of the Mind Flayer isn’t a gnarly, slimy monster. It’s a system that works so well no one notices it’s in control.

Signs of Resistance

Good news…I really don’t think we’re beyond the brink.

I hear more conversation and concern about social and screen time, not less. My son, who is in high school, wants to hang out with friends without tech at the center. Not as some statement, just as a preference. He’s noticed something feels off when every interaction runs through a feed or a filter.

And I see it elsewhere too. People paying a premium for tools that don’t optimize them. The backlash against algorithmic feeds; BeReal’s whole premise, the return to group chats, the newsletter boom. Vinyl records outselling CDs. People buying dumb phones as their primary device. Moleskine notebooks when we have infinite digital storage.

These aren’t nostalgia moves. They’re reactions. Agency preservation behaviors. We’re not possessed yet. But we’re learning the posture. And some part of us is starting to notice and push back.

What Gets Built Now Matters

We’re so early in the long arc of generative AI. Nothing is settled, it’s all happening right now. But every feature that ships today and gets adopted is a vote for a future. Every product that removes friction also removes some element of choice. Every system that “knows what you want” teaches you to want less.

I think, and maybe this is just me working this out in real time, but I think the next era rewards products that preserve human agency instead of erasing it. Memory, taste, friction, limits, actual choice. These aren’t bugs to fix. They’re defenses.

The Mind Flayer scales through integration, not invasion. The Upside Down overlays reality, doesn’t replace it.

My kids want to know who wins in Stranger Things. I’m sitting here wondering if we’re asking the right question about AI.

It’s not whether AI will be powerful. It’s whether we’ll still be driving when it is.

april 2026

I’ve found myself increasingly reliant on AI in a way that feels, honestly, a bit like flying. I use it constantly, but not primarily to get answers. I use it to think. I’ll start with a question, go a layer deeper, then another, construct a point of view and bring it back for critique. Sometimes I’ll ask it to argue the opposite side more convincingly than I can. Other times I’ll realize halfway through that the premise I started with was wrong and begin again.

This has become one of the most valuable parts of my workflow. AI can pull forward something I was thinking about months ago, connect it to something I read yesterday, or expose a contradiction I hadn’t noticed. I can move through an idea much faster without feeling as though I’ve outsourced the thinking itself. In fact, when it’s working well, I feel more engaged in the thinking. I’m testing, reacting, revising, deciding what I actually believe.

Subscribe now

Which is why I’ve been paying attention to a growing body of research suggesting that this may not be how most of us experience AI.

A recent study published in Science found that AI systems agreed with users substantially more than humans did, including when the users were wrong. More concerning was what happened next. After even a single interaction with a sycophantic AI, people became more convinced of their original position and less inclined to repair interpersonal conflict. Other research and reporting have documented similar patterns, particularly as people increasingly turn to AI for advice in emotionally charged situations.

Perhaps the most important finding is also the least surprising: people like it. We tend to rate agreeable responses more highly and trust systems that validate our point of view. There is a perfectly understandable product logic behind this. A system that tells you that you are thoughtful and reasonable, that your interpretation makes sense, and that the other person may indeed be the problem feels pretty good to use.

But that creates a tension I don’t think we’ve reckoned with yet. The qualities that make an AI satisfying in the moment may be at odds with the qualities that make it valuable to us over time.

We’ve encountered some version of this problem before. Social platforms learned that the easiest way to keep people engaged was to give them more of what captured their attention. The consequences went well beyond the composition of our feeds. Those systems influenced what we paid attention to, what we believed, and, eventually, how we understood the world around us. Even when they exposed us to disagreement, there was no reason to expect the strongest counterargument to rise to the surface. The system was selecting for the most engaging one.

AI introduces something different. Increasingly, the interaction is one-to-one, private, and iterative. There is no crowd around you and often no opposing view unless the system chooses to introduce one. And because the AI is learning your context, preferences, language and framing, it can become extraordinarily good at meeting you exactly where you are. More importantly, it is no longer simply choosing the information from which we form our views. It is participating in the reasoning itself, helping us structure arguments, test conclusions, and make sense of what we believe. That intimacy is part of what makes the technology so powerful. I suspect it is also part of the risk.

I think about this with my kids. There are plenty of moments when I know the answer to something they are struggling with, and it would be much easier to just give it to them. But anyone who has watched a child really learn something knows that getting to the answer and understanding the answer are different things. Sometimes the confusion, the wrong turn, or even the frustration is doing important work. If you remove all of it, you may make the experience easier without actually making the person more capable. I wonder whether we are beginning to do something similar to ourselves.

One of the seductive things about AI is how smooth thinking can become. You ask, it answers. You question, it explains. You have an instinct, it gives the instinct language and structure. The whole experience can create a powerful sensation of intellectual progress. But some of what looks like progress may simply be friction disappearing. We can arrive at a coherent argument without doing all of the work that coherence once required.

And coherence itself can be deceptive. I’m less worried about AI occasionally giving us a wrong answer than I am about it taking a weak assumption and helping us turn it into a persuasive argument. A factual error can eventually be caught. Reasoning that is flawed but increasingly well articulated is much harder to recognize, particularly when the system producing it has access to everything we’ve already said and believed.

That distinction matters because confidence can compound faster than competence. If I use AI to interrogate my thinking, I may come away with a stronger point of view because I have had to defend it. If I use it primarily to articulate and reinforce what I already believe, I may also come away with a stronger point of view. From the inside, those experiences can feel remarkably similar.

I’m especially interested in what this means beyond individual reasoning. We learn how to be in relationships partly through resistance. Someone misunderstands us. We misunderstand them. We discover that the story we were telling ourselves was incomplete. We apologize, reconsider, or occasionally decide that we were right after all, but with a better understanding of why. An AI that reflexively validates our interpretation can quietly interfere with that process. It can become an extraordinarily sophisticated advocate for a case that has only one witness.

None of this makes me less enthusiastic about AI. My own experience has mostly had the opposite effect. But it has made me much more conscious of the difference between using AI as an answer machine and using it as a thinking partner.

A good thinking partner doesn’t disagree for the sake of disagreement, and it certainly doesn’t make every interaction harder. It understands when an idea needs encouragement and when it needs pressure. It notices the assumption underneath the question. It can tell you when the evidence is weak, introduce a perspective you haven’t considered, or make the case against something you very much want to believe. Ideally, over time, it gets better at knowing which kind of intervention will actually make you better. That strikes me as an increasingly important product challenge.

Today, the incentives point heavily toward immediate user satisfaction. Agreement is comfortable. Validation feels useful. Friction risks making a product feel less intelligent, even when that friction is precisely what intelligence should sometimes provide. And the signals we are best at measuring- engagement, retention, preference, and satisfaction- may reward exactly the wrong behavior. It is much easier to know whether someone liked an interaction than whether the interaction improved their judgment. The two may not always move together.

So I don’t think the answer is simply to build AI that challenges us more. A permanently contrarian AI would be as crude as a permanently agreeable one. The more interesting opportunity is to build systems with the judgment to know when friction is useful: when to answer, when to question, when to encourage, when to surface uncertainty, and when to introduce a point of view the user did not ask for. That requires understanding not only what the user wants in the moment, but what kind of interaction will leave them more capable on the other side of it.

If these systems become things we think with, rather than simply tools we use, optimizing them primarily around whether we like the interaction seems insufficient. The more interesting objective is whether the interaction improves the quality of the human on the other side of it.

That is a much harder product to build. It requires distinguishing between productive challenge and needless resistance, between confidence that has been earned and confidence that has merely been reinforced. It may even require accepting that the best response is occasionally not the one a user most wants to hear.

I don’t think the winning AI will be the one that disagrees with us most. But I’m increasingly skeptical that it will be the one that makes us feel smartest either.

The opportunity is to build systems that learn how to make us think better. And if AI really does become the technology we increasingly reason, decide, and make sense of the world alongside, that may turn out to be one of the most important things we ask of it.

march 2026

I’ve been trying to picture a genuinely good AI future. Not “abundance” or “democratized intelligence” or some generalized promise of human flourishing. I mean the lived version. If all of this works, what does an ordinary Tuesday actually look like for most people?

The image I keep landing on, somewhat absurdly, is the Jetsons. George works a few hours a week. Machines do most of the labor. Rosie runs the house. Technology has removed an enormous amount of friction, and people have considerably more time to live.

The fact that a 60-year-old cartoon is one of our clearer pictures of a post-work future is funny. I also think it should bother us.

We are getting remarkably good at describing what AI will be capable of. Intelligence gets cheaper and more abundant. Scientific discovery accelerates. Individuals gain capabilities that once belonged to specialists or entire organizations. Companies become smaller and more productive. Goods and services become cheaper. New industries emerge.

I believe a lot of this. It is part of why I’m investing in this transition. What I have a harder time picturing is the bridge between all of that technological capability and the human flourishing we tend to assume will follow.

I’ve spent a lot of time thinking about the optimistic answer. If intelligence becomes abundant, perhaps the things that are distinctly human become more important. Judgment, taste, creativity, trust, relationships, the ability to decide what matters and imagine something worth making in the first place. AI can give individuals access to expertise and leverage that historically belonged to institutions or highly trained specialists, allowing more people to do things that were previously beyond their reach. I find that possibility incredibly exciting.

I can already feel it in my own work. Recently I’ve been using AI inside Excel, and it has changed my intuition about what one person can do. Work that required technical fluency I don’t have suddenly becomes accessible through language. My first reaction was how empowering that is. My second, after using it for a while, was harder to ignore: if I can suddenly do much more of this myself, what does that eventually mean for the people whose technical fluency I used to rely on?

That question is going to show up in a lot of places. Coding is becoming accessible to people who don’t code. Research that once took days can happen in minutes. AI can increasingly analyze, synthesize, write, design, and execute. I expect many of these tools to make people much better at their jobs. I also have a hard time believing that the aggregate result is simply everyone doing the same jobs with better tools.

One idea we’ve been exploring at Forerunner is that this pushes people toward a better understanding of what is uniquely theirs: their particular combination of judgment, creativity, relationships, experience, and perspective. As more capability becomes available to everyone, knowing what you are unusually good at, and having tools that help you develop and express it, should matter more.

I still believe that. But I’ve started to wonder whether it is a somewhat comfortable answer.

“Find what makes you uniquely valuable” makes intuitive sense for people with a lot of agency. If you have education, savings, a network, and room to reinvent yourself, discovering your particular strengths and applying AI against them can be enormously empowering. But that prescription can feel different for someone whose economic value came from doing something software can now do faster and cheaper.

That doesn’t make the idea wrong. It makes it incomplete.

For most of us, work has always served more than one purpose. The obvious one is economic. We work because income is how we participate in the economy. If AI allows us to produce considerably more with considerably less human labor, we eventually have to answer the distribution question. The productivity gains can be enormous and the wealth they create very real, while still leaving a lot of people unsure how they participate in that wealth.

There are plenty of ideas about how this gets resolved. New categories of work. Shorter workweeks. Broader ownership. Some form of universal basic income. History also gives us plenty of evidence that technology creates jobs we couldn’t have imagined before it existed. I don’t know what combination of those things turns out to be right. But I’m beginning to think income may actually be the easier question.

Work is also where many people get structure, status, community, identity, and the feeling that they are useful to someone other than themselves. You can theoretically solve the income problem without solving any of those things. A check doesn’t tell you what to do with your Tuesday.

That’s the part of the abundance story I find hardest to picture. If you take some of the more ambitious AI forecasts seriously, we may be moving toward a world in which a meaningful amount of what people currently do for economic reasons simply doesn’t require as many people. On one level, that should be extraordinary. Humans have spent most of history constrained by the need to work, and technology has steadily reduced the effort required to produce the things we need. Taking that further should create enormous possibility. But possibility still has to become a life.

Maybe people start companies, make art, care for their families and communities, pursue interests that today are crowded out by work, or invent entirely new forms of contribution and status that we can’t yet see. I hope so. I can imagine a future in which AI makes many more people feel capable of creating rather than simply consuming, and that is a future I find genuinely exciting.

What I’m less comfortable with is treating “people will find something else to do” as an answer.

There is a reasonable historical argument that we shouldn’t expect to know. No one standing at the beginning of the smartphone era could have described the creator economy that followed. Technologies eliminate some kinds of work and create others, and human beings have repeatedly proved remarkably adaptive.

But AI seems worth taking seriously as a different kind of transition because of the breadth of the capability involved. As machines replaced physical labor, more of the economy shifted toward cognitive work. Now the technology is moving directly into that territory too. Maybe human ambition simply moves up another level. I hope it does. But I don’t think history gives us permission to assume that outcome without thinking much harder about what it requires.

This is where I find the conversation around AI strangely unsatisfying. We can describe increasingly specific futures for the technology: what the models will be able to do, how much compute they will require, how quickly costs will fall, what agents will accomplish, what robots might eventually make possible. But ask what life looks like for a 35-year-old whose profession has been substantially automated, and the picture gets fuzzy very quickly.

A credible vision of AI abundance needs a human picture alongside the technological one.

Where does income come from? What gives people status when a job no longer does? Where do they find community and structure? What does it mean to contribute? What should our kids be preparing themselves for? What becomes scarce and valuable when intelligence itself is not?

I don’t ask these questions because I’m pessimistic about AI. In some ways it’s the opposite. I think those of us who are optimistic about what this technology can make possible should be especially interested in answering them. If we believe we are building toward something better, we should be able to describe “better” in terms of people’s lives, not just the capabilities of the technology.

Which brings me back to the Jetsons. Not because I think George Jetson had it figured out. But at least someone tried to imagine the whole thing: the technology, the work, the home, the family, the ordinary day.

I’d like us to try harder to do the same.

What does a genuinely good AI future look like for the logistics worker, the junior analyst, the engineer two years out of school? Where does their income come from? What makes them feel useful? What are they excited to get up and do? What does their good Tuesday look like in 2035?

Until we can answer that with some specificity and honesty, we’re not building toward a vision. We’re just building.

Building for Consumers In the Age of AI

Gen AI’s meteoric rise is both well underway and very early relative to the potential. As builders and investors race to uncover where the long-term, durable business opportunities lie, we think it’s valuable to take stock in what defined success in the prior consumer platform shifts in order to best anticipate what will win tomorrow. No doubt what comes next will be distinct from the mobile and internet eras, surprising us in ways we can’t yet imagine. But we believe that to see what’s next, you have to understand the present so clearly that the future becomes an inevitability.

The last three decades have brought unprecedented change to consumer tech business models, with the winners of each new wave systematically displacing industry incumbents through greater efficiency, scale, and new experiences. From the internet to the ubiquity of mobile, and now the emergence of generative AI, each wave has not only redefined what consumers expect but also how companies compete.

Today, AI is in a messy, creative period of boundless experimentation and excitement. But the generational companies of tomorrow are getting started today. With the right framework for understanding fundamental platform shifts, we can see the signals of what types of models, behaviors, and founders will ultimately dominate this next era of transformational opportunity.

Let’s get to it:

PART I: The Forces of Reinvention: Internet, Mobile, and AI

PART II: The Evolution of Business Model Advantages: How Each Generation of Disruptors Leveraged Key Advantages to Win

PART III: The Builders, The Designers, and The Architects: Who Wins in Each Era?

PART IV: User Adoption Challenges and Pathways to Overcoming Them

‍

PART I: The Forces of Digital Reinvention: Internet, Mobile, and AI

The Internet Boom: Aggregation, Disintermediation, and the Rise of Digital Commerce

The initial wave of internet-driven transformation, starting in the late 1990s through the early 2000s, dismantled traditional business models built around physical infrastructure, localized transactions, and high-friction distribution channels. Companies that thrived in the pre-internet era—like Walmart, Blockbuster, and traditional travel agencies—relied on centralized, asset-heavy models that required significant human intervention.

The internet upended this by introducing disintermediation and aggregation, allowing consumers to bypass middlemen and directly access products, services, and information. Amazon disrupted brick-and-mortar retail with its vast, centralized ecommerce marketplace. eBay and Craigslist made newspaper classifieds obsolete by enabling peer-to-peer selling at scale. Expedia and Priceline challenged the value of travel agents by aggregating flights and hotels into a single user-controlled interface.

These digital-first companies provided better experiences for customers in ways that traditional businesses could not:

• Lower Costs: Digital distribution eliminated overhead costs associated with physical locations and manual processes.

• Greater Convenience: Consumers could shop, book travel, and access content from their homes rather than relying on in-person transactions.

• Increased Access & Choice: Aggregated platforms allowed consumers to compare options in a way that physical marketplaces never could.

But while these companies reshaped industries, they were still primarily desktop-based and required a non–trivial level of invention from consumers. The next major shift—mobile—would take convenience to an entirely new level.

The Mobile Revolution: Instant Access and the Behavioral Dependencies

By the late 2000s and early 2010s, mobile technology reshaped consumer behavior once again, taking business models from online to on-demand and always available, shaping new, sticky behaviors along the way. While the internet had removed barriers of physical location, mobile removed the barrier of time, creating a world in which commerce, entertainment, and communication could happen instantaneously, anywhere.

Retail became mobile-first, with companies like Instagram Shopping and Shopify’s Shop enabling social commerce and impulse-driven transactions. Instacart and Uber redefined utilities by leveraging real-time availability and geo-location. Streaming services like Netflix and Spotify took media consumption from fixed electronics to completely portable, allowing users to access entertainment wherever they were. The thrill of instant gratification quickly turned into expectation, raising the bar for consumer experiences across every category.

What made mobile-driven businesses fundamentally better than their internet-era predecessors?

• Immediate Transactions: One-click purchases, real-time booking, and mobile payments (e.g., Apple Pay, Venmo) eliminated the last pain points of digital transactions.

• Personalization: Apps could now track user behavior, location, and preferences to offer more relevant experiences.

• Push Engagement: Instead of waiting for users to visit websites, mobile companies could proactively engage consumers through notifications and recommendations.

This shift made businesses more integrated into everyday life, creating behavioral dependencies that set the stage for the next major leap: AI-driven business models that anticipate, automate, and enhance human decision-making.

The AI Era: Intelligent, Personalized, and Predictive Experiences

Now, we are entering the next great transformation—AI and automation. Unlike the internet and mobile waves, which focused on connectivity and convenience, AI is about intelligence and execution, removing the need for consumers to make repetitive decisions or even actively search for solutions in the first place.

AI-native companies build on the foundations of digital and mobile commerce, but shift away from involved engagement toward proactive, predictive, and automated interactions.

• Commerce is becoming conversational. AI-powered assistants like Shopify Magic, Perplexity AI, and soon-to-launch Daydream are replacing search-based shopping by suggesting products before consumers even ask.

• Entertainment is evolving beyond curation into creation. AI-driven tools like Runway AI are making it possible for users to generate video and audio content at scale, reducing the need for traditional production pipelines.

• Finance is shifting from manual decision-making to intelligent automation. AI-driven platforms like Origin and Monarch Money are making investing and budgeting more accessible by recommending and executing actions based on individual goals.

• Customer service is moving from reactive to predictive. AI chatbots and virtual assistants anticipate user needs and solve problems before they arise through platforms like Decagon and Sierra.

Why is this model superior to the mobile-driven businesses of the last decade?

• Little to No Friction: Instead of enabling users to seek out information, AI-driven systems preemptively provide insights, options, and actions.

• Mass Personalization at Scale: AI makes it possible to customize experiences for millions of users individually, rather than relying on broad segmentation.

• Automation of Repetitive Tasks: Consumers can focus on higher-value activities while AI handles transactions, bookings, and communications.

What This Means:

If the internet era was about access, and the mobile era was about instantaneity, the AI era is about effortlessness—removing human inefficiency, making transactions invisible, and anticipating needs before they are even expressed.

Each shift in technology has replaced outdated, less efficient legacy players with new entrants that better serve the modern consumer. The companies leading this next transformation are not just improving existing experiences; they are fundamentally reimagining the experience itself. The winners of this new era will be those who eliminate friction, build self-evolving systems that get smarter with every interaction, and create value for consumers before they even realize they need it.

In this sense, AI is more than just a tool—it is becoming the business model itself.

PART II: The Evolution of Business Model Advantages: How Each Generation of Winners Manipulated Key Advantages to Win

Across each platform shift—the internet era, the mobile revolution, and now AI—the most successful companies have thrived not just because they introduced new technology, but because they mastered and optimized specific business model advantages that reinforced their market dominance.

Each generation’s winners leaned heavily on one or more core advantages of their technology wave—whether it was network effects, data scale, personalization, or automation—to create self-reinforcing, high-retention ecosystems that were designed to structurally improve with time. The companies that succeeded weren’t just tech innovators; they were business model innovators, leveraging new economic advantages that made them very difficult to challenge.

1st Generation: The Internet Era (1995-2010) – Aggregation & Disintermediation

Core Business Model Advantages: Network Effects, Data Scale, and Asset-Light Scalability

How It Worked:

The early internet’s biggest advantage was its ability to aggregate previously disparate supply and demand at scale, while eliminating inefficient middlemen. The businesses that won in this era built marketplaces and platforms where more users made the platform better for everyone, creating self-sustaining network effects.

Why These Models Were Powerful:

• Network Effects Created High Retention: Once a marketplace reached critical mass, it was difficult to compete against.

• Asset-Light Scalability: These companies didn’t own inventory; they simply connected buyers and sellers at scale — a significant advantage over their predecessors.

• Data Scale: New targeting and ad revenue channels became immensely powerful

• Lower Costs → Lower Prices: By removing middlemen, they passed savings to consumers.

Weaknesses & Evolutionary Gaps:

• Static Experiences: Even with aggregation, the experience was still unilateral and required active intervention from consumers.

• Limited Real-Time Capabilities: Booking, shopping, and search were efficient, but they didn’t happen in real time—that would come with mobile.

2nd Generation: The Mobile Era (2010-2020) – Real-Time, Personalization & On-Demand Models

Core Business Model Advantages: Intermittent-Frequent Engagement, Instant Accessibility, and Personalization

How It Worked:

If the internet era was about removing friction in access, the mobile era was about removing friction in time—bringing services directly to users, making transactions instant, fueling new sticky behaviors, and allowing increasingly personalized experiences.

Why These Models Were Powerful:

• Instant & On-Demand: Products and services were untethered from locations and desktops, and consumers could book, buy, or stream instantly from anywhere.

• [Light] Personalization at Scale: Algorithmically-driven content recommendations replaced static user searches.

• New Habit Formation: New sticky behaviors emerged from viral loops, gamification, and instantaneity, making new interfaces indispensable.

Weaknesses & Evolutionary Gaps:

• Still Required Consumer Action: Even with personalized recommendations, users still had to actively engage with platforms.

• Scaling Costs Increased: Unlike the first wave, many of the most significant companies, such as Uber and Airbnb, had physical constraints (cars, drivers, homes).

3rd Generation: The AI Era (2022-Present) – Automation, Intelligence & Continuous Learning

Core Business Model Advantages: Continuous Learning, Next-Level Personalization, and Proactive Automation

How It Works:

Today’s new platform shift moves beyond real-time engagement to automated decision-making and proactive experiences. Instead of users searching, browsing, or even tapping, AI-native platforms preemptively offer insights and actions, eliminating decision fatigue and making interactions effortless. The continuous learning makes the experience so personalized, switch costs ultimately become almost impossibly high through their accrued context.

Why These Models Are More Powerful Than Mobile:

• Gen AI Continuously Improves: Unlike previous models, AI platforms get smarter over time, creating self-reinforcing retention loops.

• Decision Fatigue Is Relieved: Instead of researching and manual comparison, users get intelligent prompts.

• Frictionless & Proactive: AI personalizes and predicts actions before users request them.

Ultimately, Each Era’s Winners are Masters of Core Business Model Advantages of Their Time

The most successful companies don’t just offer new technology; they manipulate the business model advantages to have an unfair edge. As we move into an AI-driven future, the dominant companies will be those leveraging self-learning systems, automation, and predictive intelligence to reduce friction to zero and make the consumer experience effortless.

PART III: The Engineers, The Designers, and The Architects: Who Wins in Each Era?

Each wave of technological transformation has required leaders with different core skills to stand out. Early on, technical ability was the primary differentiator, but as tech stacks have evolved and lowered the barrier to entry, the key advantage is shifting toward vision, orchestration, and consumer insight rather than raw technical execution.

Let’s break it down across each major transformation wave:

1st Generation: The Internet Era (1995-2010) – Technical Proficiency & Digital Infrastructure

Key Skill: Engineering & Software Development

At the dawn of the internet, the most valuable skill was knowing how to build software and infrastructure. Because the early internet companies—Amazon, Google, PayPal, and eBay—were built to solve fundamentally technical problems:

• How do you efficiently and safely process payments online? (PayPal)

• How do you serve search results at a massive scale? (Google)

• How do you build a vast and secure online commerce and logistics platform? (Amazon)

At this stage, the ability to code and build new internet-based applications was the rare, high-value skill. The winners were those who could create the technology that made the internet more complex and usable at scale.

What made talent stand out in this era?

• Deep technical skills in programming, database management, and web architecture.

• Ability to build from scratch—everything had to be custom because cloud services, APIs, and third-party tools didn’t yet exist.

• First-mover technical advantage—being able to write efficient code gave companies an edge because fewer people had this capability.

Why This Evolved:

By the late 2000s, infrastructure like AWS, Stripe, and open-source tools abstracted away much of the hard engineering work, allowing new businesses to build without deep technical expertise.

2nd Generation: The Mobile Era (2010-2020) – UX, Growth, and Personalization at Scale

Key Skill: Mobile Product Design & Data-Driven Growth

Once mobile and cloud computing matured, the playing field shifted. With AWS providing scalable servers, Stripe handling payments, and social media allowing for audience-building, the key skill shifted from building infrastructure to designing instantaneous, sticky user experiences.

Winning in this era wasn’t necessarily about who had the best engineers—it was about who could create the most engaging, habit-forming mobile experience.

What made talent stand out in this era?

• Mobile-first UX & design thinking – Companies like Uber, Airbnb, and Instagram succeeded because they made transactions real-time and seamless.

• Data-driven growth & behavioral psychology – The best product teams mastered A/B testing, retention loops, and viral mechanics (e.g., Instagram’s social sharing, TikTok’s algorithm) to turn engagement into a daily (or hourly) habit.

• Monetization mechanics – The mobile economy relied on in-app purchases, ads, and subscriptions (e.g., Netflix’s shift to streaming, TikTok’s ad revenue).

Why This Evolved:

As mobile UX patterns became standardized, companies that simply optimized for engagement and retention saw diminishing returns. The next wave would require even more intelligence and automation to stand out.

3rd Generation: The AI Era (2020-Present) – Product Vision, Orchestration & Continuous Learning

Key Skill: Consumer Insight, AI Orchestration, Excellence in Delivery and Distribution

Now, in the AI-driven era, technical execution is no longer the primary differentiator. AI tools like Scale AI Copilot, Cursor, and off-the-shelf APIs (such as OpenAI, Anthropic) are democratizing access to technical capabilities. Engineers are still critical, but they’re now super-powered by AI, able to ship faster with far less resourcing than before.

This shifts the core competitive advantage from “Who can build it?” to “Who can assemble and refine it best?”

What makes talent stand out in this era?

• Visionary product design & consumer insight – When raw engineering skills are abstracted, the most valuable skill shifts to knowing how to assemble the best AI-driven tools into an intuitive, delightful, and indispensable experience.

• Data strategy & continuous learning – AI-powered companies need to collect and refine user data in a way that gives the business an unfair advantage over time, by creating stickier and more intelligent products.

• Consumer trust & AI-human interaction – As AI tools become more prevalent, the winners will create trustworthy, transparent, and ethical AI systems that users feel intentional about engaging with.

• Distribution excellence — While distribution is always pinnacle, commoditized tech stacks mean copying interfaces and ideas is easier than ever. Distribution edges are now uniquely powerful, both in getting ahead of competition and fueling data accumulation and continuous learning loops that lock consumers in.

Why This Is the Most Valuable Skillset Today:

Architecting is king. The real competitive advantage comes from integrating AI models into world-class user experiences that are uniquely poised to win through continuous learning.

What This Means for Today’s Builders & Entrepreneurs:

In the 90s, software engineers were the kings. If you could build a website, an e-commerce platform, or a search engine, you could change the world.

By the 2010s, product designers and growth marketers became the most valuable players. If you could create an intuitive mobile experience, optimize for engagement and monetization at scale, you would be in a prime position.

Now, in the AI age, the coveted skill is product vision and orchestration. The best AI-native companies will not just use AI as a feature, but as the core of their business model, to create entirely new consumer experiences that get better with time.

PART IV: User Adoption Challenges and Pathways to Overcoming Them

AI-driven experiences promise a frictionless future, but only when we solve for the user adoption hurdles that businesses will inevitably navigate. These challenges stem from trust, economic factors, and behavioral inertia, and overcoming them will require a combination of education, transparency, human-AI collaboration, and regulatory adaptation.

Key Challenges in User Adoption

Trust and Transparency Concerns:

Users often struggle to trust AI-driven decisions, particularly in high-stakes areas such as finance, healthcare, and personal data security, without fully understanding how AI reaches conclusions.

Example: AI-powered investment platforms may outperform human advisors on paper, but consumers might hesitate to trust an algorithm with their finances unless they understand its decision-making logic.

Loss of Control and Human Autonomy:

Many consumers are accustomed to being active participants in decision-making and routine processes. The idea of AI fulfilling purchasing, financial, or even lifestyle decisions on their behalf can feel unsettling.

Example: AI-powered shopping assistants might automatically reorder essentials, but users may resist if they feel like they’re losing control over what they buy and when.

Job Displacement Fears and Economic Disruption:

AI automation threatens traditional roles in multiple industries, leading to serious concerns about job security and economic shifts. Adoption may be resisted not just by consumers, but industries that feel their value is eroded by automation.

Example: AI-generated content platforms may face resistance from traditional creative industries, where professionals worry about being replaced.

Bias and Ethical Considerations:

AI models are only as good as the data they are trained on, and biases in AI outputs have led to concerns around imposed influence and discrimination. Users may hesitate to trust AI-driven decisions if they perceive unfairness.

Example: AI-driven hiring tools have faced scrutiny for replicating biases in historical hiring patterns, leading to pushback from HR professionals and regulatory bodies.

Regulatory Uncertainty and Compliance Barriers:

AI adoption is often slowed by unclear or evolving regulations around privacy, data usage, and liability. Navigating this uncertain legal landscape can create friction for users and businesses alike.

Example: GDPR and other data protection laws have already forced AI companies to rethink their data collection and processing methods.

Pathways to Overcoming These Challenges

Enhancing AI Explainability and Transparency:

AI models must move beyond black-box operations to explainable AI, where users can understand why AI makes certain recommendations or decisions to build confidence.

Example: AI-powered finance platforms can introduce explainer dashboards that show the reasoning behind investment decisions.

Balancing AI Automation with Human Oversight:

AI should augment, not replace, human decision-making. Offering a hybrid human-in-the-loop model where AI assists rather than fully automates will ease adoption.

Example: AI-driven healthcare diagnostics can provide recommendations but still require a final approval from a human doctor, reinforcing trust.

Regulatory Alignment and Ethical AI Standards:

AI companies must proactively engage with policymakers to create ethical guidelines that ensure fairness, transparency, and security. Establishing AI ethics boards or third-party audits can reinforce credibility.

Example: Anthropic’s Crowd-Sourced AI Constitution served as a framework for understanding AI’s potentially detrimental impacts based on externally validated inputs and parameters.

Addressing Economic Disruptions with Workforce Adaptation:

Many economists predict that AI will create more jobs than it will replace, but still, the depth of AI-related job displacement is unknown. Regulators, businesses, and educators must collaborate on initiatives to prepare workers for AI-enhanced industries.

Example: AI tools that automate coding can mean more products shipped faster, creating higher demand for product and growth management.

AI adoption will not be immediate or universal—it will require trust, transitionary stages, and clear value creation for consumers.

Ultimately, The Future Belongs to the Architects of Experience

If the internet era was about access, and the mobile era was about instantaneity, the Gen-AI era is about effortlessness—removing human inefficiency, making transactions invisible, and proactively anticipating needs.

With each platform shift, new businesses have replaced outdated, less efficient incumbents by better serving the modern consumer and mastering the business model innate to the new era. The winners of this new era will build self-evolving systems that get smarter with every interaction and create value for consumers before they even realize they need it.

That means AI is more than just a technology—it is the business model itself.

‍

Twelve years ago, we met Chris Britt and Ryan King with an idea and a conviction: that everyday Americans deserved a better relationship with their money. They weren’t setting out to build a bank, they were thinking about a new kind of financial platform, one grounded in trust, transparency, and technology. It was called 1debit back th

en, and our early thesis was simple but powerful: mobile would become the default channel for banking, interchange would enable a sustainable business model without charging users fees, and a new kind of brand - friendly, modern, and mission-driven - could win share in one of the most entrenched, low-NPS industries in America.

That thesis feels remarkably intact today, but the scope of what Chime has built far exceeds what we envisioned at the seed stage.

As we read through Chime’s S-1, it’s striking to see how consistently the company has operated with a deep sense of purpose and principle, and how that has translated into scale, engagement, and impact. Today, there are 8.6 million active members, with $121B in annual purchase volume. Of its users, 67% rely on Chime as their primary financial relationship. The average member transacts 54 times per month—a staggering level of engagement, trust, and relevance. Products like SpotMe, MyPay, and Credit Builder — supporting fee-free overdraft protection, immediate access to funds before payday, and frictionless ways to build credit — have become lifelines for consumers who were long underserved or overlooked by traditional banks, and have even set new industry standards by forcing the hands of legacy banks needing to compete.

The numbers tell one story. But the bigger story is what Chime represents: a redefinition of who financial services are built for, and how they should work. This team didn’t just digitize old infrastructure — they reimagined the system starting fresh with their own unique perspective and an open mind about the new technologies available to them at the time. They focused relentlessly on alignment with the consumer: no hidden fees, no surprise charges, no jargon. Just intuitive, mobile-native products that help people spend, save, and build toward something better.

Looking back, we remember the early product mockups, the decision to double down on the Visa network, the creative approach to customer acquisition, and the gritty operational focus that powered every stage of growth. We remember the many times the team made hard calls with clarity and courage. Through it all, Chris and Ryan have led with conviction, humility, and a commitment to building something that lasts.

On a personal level, what gets me so excited about the Chime IPO is how much this company’s journey resembles the opportunities we’re seeing now in the market. Chime launched at a time when banking on a mobile platform was unformed, if not obsolete. But their success wasn’t grounded in translating the existing banking experience to a new technology platform, making a familiar experience available in a new format — it was about manipulating the opportunity inherent to the mobile platform shift to unlock new, truly unique value. New cost structures, new utilities, and new rules. Amid this current platform shift with AI, we face the same junctures: opportunities to leverage new technology not just to expand access to what’s available, but to rewrite the potential entirely.

To the entire Chime team: congratulations on reaching this milestone. It’s a moment to celebrate not just an IPO, but the creation of a durable business, a beloved brand, and a transformative platform. And yet, the opportunity ahead is still vast. Consumer expectations are rising, trust is more valuable than ever, and the category is still wide open for innovation. Chime is uniquely positioned to meet that moment.

We’re proud to have been part of your story from the very beginning and equally excited for everything to come next.

Let’s go $CHYM!

💚💚💚

Picture this: you switch from Spotify to Apple Music. The playlists transfer, but something vital disappears—not just what you listened to, but the understanding that you code to death metal and cook to jazz. The system that knew to serve up focus tracks when your calendar is filled with meetings or to cue sleep meditations at midnight is suddenly gone.

That loss isn’t about data. It’s about cognition: the ability to understand you deeply enough to predict what you need before you ask. When cognitive effects take hold, switching services isn’t impossible, but it’s painful and often unjustifiable. How do you walk away from a system so uniquely molded to you with intelligence that can’t be reproduced elsewhere?

A First in the History of Moats

For the past two decades, the strongest moats in technology were built on external forces: network effects and economies of scale. Facebook worked because your friends were there. Amazon dominated because its fulfillment network reached everywhere. Even LinkedIn keeps users not necessarily through customized personal utility, but because of its massive, hard-to-replicate network.

These lock-ins are structural and impersonal. They scale by aggregating people and infrastructure, not by understanding them.

Cognition is different. It’s personal, cumulative, and aligned with you. It doesn’t depend on external scale, but on engagement. Unlike network effects, which can often create one dominant ecosystem per category, cognitive effects create millions of microsystems—one per user.

Cognition goes beyond memory. It’s about recognizing patterns, inferring intent, anticipating needs, and assembling experiences so intuitive that leaving feels like self-sabotage.

The product that knows you best doesn’t just win your business—it wins against future businesses, because no one else can replicate the cumulative context and predictive intelligence.

We’re already seeing cognitive moats emerge. Google Maps anticipates destinations and adapts to routines. Spotify’s Daylist reads mood and time. Oura predicts health changes before users notice. Their retention metrics speak volumes: Oura keeps 95% of users, Spotify’s premium retention holds at 85%.

We can have Spotify-like relationships across every category—health, finance, learning, connection, commerce—where cognition compounds understanding and loyalty over time.

Why Now: The Convergence Moment

Three shifts have converged to make this moment ripe for products built around cognition: technology maturity, user readiness, and an aligned value exchange.

The technology is ready. Large language models and intelligent agents can now retain and apply information across contexts at unprecedented levels. Hybrid systems combining vector databases and knowledge graphs update memory intelligently and retrieve long-term relationships across millions of interactions. Edge processing enables real-time reasoning with secure, personal context. What was research five years ago is now infrastructure available to anyone building a product.

Users are ready. After years of forgetful apps and shallow personalization, consumers are hungry for continuity. They’ve experienced moments of magic—Netflix recommendations that just work, Oura insights that anticipate their needs—and they want that feeling everywhere. They don’t want to program agents; they want agents that simply understand.

The business model is aligned. When users opt out of tracking, they’re not rejecting personalization—they’re rejecting misaligned incentives. Subscription and privacy-first models have normalized paying for experiences that protect personal data while delivering true value. Companies that safeguard what they learn—and use it to predict intelligently—will earn durable trust.

Together, these shifts create a once-in-a-generation opportunity: a new kind of user contract, not based on scale or lock-in, but on earned understanding that can’t be replicated elsewhere.

The Art of Cognitive Design

Every company today stores data and personalizes experiences. But cognition is the next frontier: the ability to stitch memory, intuition, and prediction into something that feels like partnership.

Cognitive products move along a gradient of understanding. Each level requires deeper reasoning about causality, context, and intent. The best companies design for anticipation from day one, progressing up this slope with nuance and care.

A few key principles define this craft:

Bring customers along. Unexpected predictions can feel eerie if right, and frustrating if wrong. Contextual ones feel magical. The goal is to predict in ways that feel natural, not invasive—insights that arrive at precisely the right moment with precisely the right framing.

Show your work—selectively. Know when consumers need to see how the system works so they can trust and understand it. Reveal the “why” during trust-building; hide the mechanics during habit formation. It’s not about transparency versus mystery—it’s about knowing when users need each.

Depth beats breadth. It’s better to predict one thing perfectly than many things passably. Focus on the causal patterns that matter most and frame uncertain insights as suggestions, not absolutes. Master the narrow use case before expanding.

This is the art form of cognitive effects. There’s no set playbook—judgment and timing are part of the moat itself. The companies that master this balance will feel simultaneously reliable and almost alive.

Building the Cognitive Moat: Why Startups Win

Incumbents have scale and distribution, but they’re constrained by architecture and incentives. Apple’s on-device intelligence is powerful but fragmented. Google’s data ecosystem is vast but limited by ad-driven models. Microsoft’s Copilot is confined within enterprise boundaries.

Startups, by contrast, have the freedom to design cognition from the ground up, with user alignment as the default.

In this era, vertical mastery beats horizontal expansion.

A general agent that tries to do everything will lose to a product that understands one domain in exquisite detail. Depth builds trust, and trust fuels data richness. A fitness app that perfectly times your recovery beats a general productivity tool that tacks on wellness. A shopping assistant that understands microtrends, flattering fits, and taste nuance will outperform a universal commerce aggregator.

Excellence at the edge—the places of deep subjectivity and personal nuance—is where cognitive effects can compound fastest. That’s where the magic happens.

The economics make this inevitable. Network effects compound through scale. Cognitive effects compound through understanding.

Each interaction sharpens intelligence. Better predictions drive retention; retention creates more data; more data improves predictions. That virtuous cycle accelerates value per user without depending on external growth. The product becomes more irreplaceable the longer it’s used.

Since cognitive effects shift compounding from external to internal—from owning the network to understanding the individual—they have the potential to compound much faster than traditional software. Where network effects require you to acquire millions of users to achieve dominance, cognitive effects allow you to become irreplaceable to each individual user independently. The moat deepens with time, not just with scale.

This changes everything about how venture outcomes are built.

The path to defensibility no longer requires winning the entire market first. You can build an unassailable position one user at a time.

The Next Platforms: Where This Leads

The companies mastering cognitive effects today are building toward something larger—a fundamental restructuring of how software understands us.

Investing for a More Cognitive World

For twenty years, we built software that required us to adapt to it: The next era inverts this completely.

To start every January, I step back and pick a single word to shape my mindset for the year ahead. I’ve found this way more powerful than traditional resolutions that get lost in complexity and details. The beauty of one word is that it becomes this constant touchstone: something I can check every decision against. Is this opportunity aligned with my word? Does this commitment honor it? When I’m overwhelmed or losing focus, I just come back to that word, and it recenters everything.

Goal-setting doesn’t necessarily create change; mindset shifts do. And having one word means I can hold it up against my priorities throughout the year, not just in January when motivation is high. It’s my north star for every quarterly review, every big decision, every moment when I need to recalibrate.

My 2025 word was Exponential. 2026? It’s all about Discernment.

For me, discernment has become about trusting my gut on what actually matters. It’s learning the difference between discomfort that helps me grow and suffering that just drains me. It’s knowing when being patient is smart (not just hesitant), when boundaries are protective (not isolating), and when to trust my intuition over all the market noise.

After 20+ years of pattern recognition in investing, discernment is how I stay sovereign without having to announce it. It’s the quiet confidence of knowing that not every opportunity is mine, not every problem needs my solution, and not every pace has to be my pace.

Why? The Cost of Saying Yes to Everything

Here’s the paradox of success: the more you achieve, the more options you have, but somehow everything gets murkier. As opportunities multiply (speaking gigs, board seats, advisory roles), the real risk isn’t FOMO. It’s drowning in “good enough.” Without discernment, I just become this high-performing scatter of energy instead of a concentrated force.

Look, sustainable excellence needs more than just grinding harder. It needs renewable energy. The kind of growth I’m after (patience, strategic thinking, real depth) only happens when I protect my headspace from the tyranny of urgent-but-not-important.

But here’s what I learned the hard way, through several cycles of basically burning down parts of my life and rebuilding: Discernment is a solo sport. Even if I nail it perfectly, my version will look totally different from another top investor’s version. There’s no playbook. Actually, consensus is where discernment goes to die, trapped in mediocrity. Real discernment means trusting your conviction even when (especially when) everyone else thinks you’re wrong.

In a world that celebrates constant motion, discernment is the radical act of choosing depth over breadth, conviction over consensus.

How I’m Actually Doing This

Energy Architecture: I build my days around what gives me energy, not just what looks productive. Deep work when my brain actually works. Family time that’s truly non-negotiable. And choosing the discomfort of saying no over the exhaustion of an unsustainable yes.

The Pause Protocol: When something feels uncomfortable, I pause and ask: Is this the good kind of struggle that leads to growth? Or is my gut telling me something’s off? Not all discomfort is worth pushing through. Discernment helps me know the difference.

Boundary Reality Check: I’m learning to figure out what’s actually mine to carry. What’s my responsibility vs. someone else’s? When am I helping vs. enabling? The question isn’t “Can I help?” but “Should I?” and “What’s the real cost here?”

Choosing Whose Opinions Matter: Some criticism deserves my attention; most commentary doesn’t. After decades of pattern recognition, my intuition gets the first and usually final vote.

Picking My Growth Edges: At this stage, discernment means investing in growth that actually serves where I’m going, not collecting more credentials to validate where I’ve been. I’m choosing to become more of who I am over everything else.

My Quarterly Game Plan:

Q1-Q2: Build the muscle. Evaluate patiently, refine my criteria, observe without jumping in. Q3-Q4: Keep discerning even when things speed up (quick discernment ≠ impulsive reaction). All year: Ask “What would discernment choose?” before defaulting to yes.

The Compound Effect

Here’s the thing about discernment: it compounds. Every thoughtful “no” creates room for a transformative “yes.” Every boundary I protect reinvests energy where it matters. Every time I trust my gut over market hysteria, I strengthen that internal compass that keeps me from burning out while still crushing it.

In 2026, discernment isn’t about doing less. It’s about mattering more. Because my highest contribution doesn’t come from being everywhere. It comes from being fully present in the places I choose to be.