Smooths and Jags

---

I want to show you one more Tuesday before we go further. Not a warning. A picture. Hold it loosely. You don't have to agree with it to see it.

It's 2046. The person waking up is someone who was eight years old when I wrote this, a child in a classroom somewhere, maybe one very much like the ones my wife spent her career in. They are twenty-eight now. They check their assistant before their eyes have fully adjusted, the same way you do now, except there is no longer a gap between the asking and the knowing, no longer a lag where the phone loads and the answer catches up to the question. The assistant has already anticipated many of the questions they'll ask before they've consciously formulated them. This isn't intrusive. It's the most natural thing in the world to them. They have never known anything else.

They get to work, if work is still the word for it, at something that requires judgment, a genuinely ambiguous situation with no clean answer, the kind of decision that used to take forty minutes of sitting with discomfort. The assistant offers a structured breakdown in four seconds instead of the four minutes I described in the first Tuesday. The options are ranked. The tradeoffs are weighted. Our twenty-eight-year-old picks one. The decision gets made.

Here's the part I want you to actually sit with, because it's the part a simple warning skips past: the decision they made was probably fine. Statistically, probably a good decision, maybe a better decision on average than the one a tired, distracted, cognitively taxed human would have made working it through alone. That's not a small thing to concede, and I'm not going to pretend it's not true just because it complicates my argument. The frictionless choice is very often the good choice, in the moment, on that specific Tuesday, for that specific decision.

What I want you to picture is not that Tuesday. It's the ten thousandth Tuesday like it. Because somewhere around Tuesday number four thousand, or six thousand, nobody's counting, a decision arrives that doesn't fit the pattern. The situation is genuinely novel in a way the assistant's training doesn't quite cover, or the assistant's confident answer is subtly, dangerously wrong in a way that only becomes visible to someone who has spent years learning to distrust confident answers on principle. And the person facing that decision is twenty-eight years into a life where the muscle for sitting inside an unresolved, ambiguous, genuinely hard question has had almost no exercise. Not because they're lazy. Not because they're incurious. Because the exercise was never asked of them, ten thousand times in a row, and the body, in this case the mind, does not build what it is never asked to use.

Picture the room where this happens. It could be a hospital, a courtroom, a cockpit, a boardroom, a kitchen table where a parent is trying to figure out whether their child's fever is the ordinary kind or the kind that needs an emergency room right now, at two in the morning, with no assistant answer that feels quite trustworthy enough. It doesn't matter which room. What matters is that the room has, for the first time in that person's professional or personal life, run out of protocol. The assistant is still there, still confident, still fast. But confidence and speed were never the thing that mattered in that particular room. What mattered was the thing underneath them, the capacity to sit inside not-knowing until something true emerged from the sitting. And that capacity, in this picture, in this one imagined life twenty years from now, was never built, because it was never once required.

I don't know if this is what 2046 actually looks like. I told you plainly earlier in this piece that I'm reporting a trajectory from inside a fog, and I meant it. Maybe the assistants get good enough that the room never runs out of protocol, ever, for anyone, and this whole scenario simply never occurs, and I will have spent a great deal of energy worrying about a Tuesday that never happens. That's possible. I don't think it's likely, because sufficiently complex systems have consistently encountered conditions nobody designed for. That's closer to a historical pattern than a hunch, but I want to be honest that I can't rule it out.

What I can offer you, instead of certainty, is a second picture, right next to the first one, because I think the comparison is the actual point of showing you either of them.

Somewhere else in 2046, another twenty-eight-year-old wakes up. Same year, similar assistant technology available to them, equally capable, equally fast. But somewhere in this person's childhood, a parent, a teacher, a mentor, made a set of small, unfashionable, occasionally inconvenient choices. Let the math problem sit unsolved for twenty extra minutes before the tool got involved. Insisted on the uncomfortable conversation instead of the seven-step framework. Chose the harder route home, sometimes, on purpose, for no reason except that choosing it was the whole point. This person also uses the assistant every day. They are not a Luddite. They are not performing struggle for an audience. They simply have, sitting quietly underneath their fluency with the tools of 2046, a capacity that got exercised often enough, early enough, that it consolidated the way capacities consolidate when they're used. When their room finally runs out of protocol, and eventually it will, because eventually every room does, they have something to reach for that isn't just a faster restatement of the question they already asked.

I am not telling you the second person is better than the first. I want to be as careful about that as I've tried to be about everything else in this piece. The first person is not lesser. They are not weak, not stupid, not a lesser form of the same species. They are the product of an environment that gave them everything except the one kind of exercise that only shows up as valuable in the rooms where the protocol has run out, and most rooms, most days, for most of a life, never run out of protocol. The first person will likely be happier, on average, across the ten thousand ordinary Tuesdays. The second person will have spent a meaningful amount of that same time in low-grade, voluntary discomfort that mostly didn't pay off in any way they could point to, because until the room runs out, the discomfort just looks like inefficiency.

That's the actual shape of the bet, as best I can construct it. Not virtue against vice. Not smart against dumb. It's efficient-until-tested against inefficient-until-needed, and the only honest thing I can tell you is that you don't get to know in advance which Tuesday is the one that tests you, or whether it comes at all.

I think about which of those two twenty-eight-year-olds I'd want making the call if it were my own diagnosis on the table, ambiguous, the scan inconclusive, the protocol silent. I think about which one I'd want on the jury if I were the defendant, the one who can hold two irreconcilable stories in their head without a summary handing them the answer, or the one who's very good, very fast, at accepting the cleanest available resolution. I think about which one is teaching my hypothetical grandchild, twenty years from now, when that lesson is one the assistant genuinely can't finish on the child's behalf, however much the child might quietly want it to.

I don't know which future actually arrives. I told you that already, and I meant it every time I've said it in this piece. What I know is that right now, this year, while the fog is still thick enough that nobody can prove me wrong or right, is exactly the moment when the choice between those two Tuesdays is still being made, quietly, in a million individual houses, by people who mostly don't know they're making it. You don't get a notification when you make this choice. There's no consent form. There's just a hand moving halfway toward the phone, this time, and then again tomorrow, and then again the day after that, ten thousand times, until one day it's simply who somebody is.

I'm not asking you to be alarmed by this. I've made my peace with the idea that alarm isn't actually what this deserves, and I'd rather you finish this piece thinking "that's an interesting way to look at it, time will tell" than thinking I tried to frighten you into a position. But I am asking you to actually hold the picture for a minute. Both Tuesdays. Both twenty-eight-year-olds. Neither one is imaginary in the sense that matters, because both are simply extensions, run forward, of choices already being made today, in your house, on an ordinary Tuesday, by you, whether or not you noticed you were making them.

That's the whole ask. Not belief. Just the willingness to actually picture it, for a minute, before you put the phone down and go on with your day.   

Here's where I have to be honest about something, because a forensic auditor who isn't honest about the weakness of his own evidence isn't an auditor. He's an advocate. The machines told me where to look. They are not why you have to believe me.

Three AI systems converging on the same mechanism under sustained pressure was, to my mind, a real and unusual thing to have watched happen. It was also the reason I went looking for evidence, not a substitute for it.

It is not, on its own, proof. A language model under sustained adversarial questioning is not a witness under oath. It's a system generating plausible continuations shaped by the conversation itself, including by me, including by the pressure I was applying. I know that. I want you to know I know that, because the people who will dismiss this page, and some of them will, are going to make that point, and they'll be making a fair point when they do.

So I went looking for the part of this argument that survives without anything a chatbot ever said to me at two in the morning. I went looking for the number that reconciles on its own, without my thumb anywhere near the scale.

I found five.

The earliest appearance was in 2011: two researchers named Betsy Sparrow and Daniel Wegner, along with Jenny Liu, published a study in the journal Science that had nothing to do with AI, because AI as we know it didn't exist yet. They found that when people expect to have future access to information, they become measurably worse at remembering the information itself and measurably better at remembering where to find it. The mind, given a reliable external memory, stops doing the work of holding the thing itself. That's not a theory about AI. That's a documented finding about what any sufficiently reliable external tool does to human memory, published over a decade before the tool got a voice and started talking back.

In 2025, a team led by researchers including Hamsa Bastani ran a randomized controlled trial with roughly a thousand high school students in Turkey, studying math performance with and without AI assistance. Unassisted exam performance measurably dropped after students had a period of unrestricted AI access during practice. Not felt worse. Measured worse. On a test taken without the tool in the room.

In 2026, researchers named Shen and Tamkin studied fifty-two professional software developers, not students, adults already established in a skilled profession, and found a seventeen percent reduction in conceptual understanding among those who used AI assistance, despite no meaningful time savings over the ones who didn't. The tool didn't make them faster. It left them with shallower conceptual understanding, and the shallowness didn't show up as slower, cheaper work. It hid inside work that looked, from the outside, exactly the same.

In 2025, a researcher named Michael Gerlich published a study in the journal Societies finding a measurable relationship between frequent AI tool use and reduced critical thinking performance, and the relationship ran specifically through increased cognitive offloading, meaning the cognitive offloading wasn't simply correlated with the effect. It statistically mediated it. It was the actual channel through which the effect traveled.

And in 2025, MIT's Media Lab ran an EEG study on participants writing essays with and without AI assistance, and found the weakest neural connectivity, measured directly in brain activity, not self-report, among the group that used AI. More than eighty percent of that group could not accurately quote from an essay they had written minutes earlier. They had produced the words. What the EEG measured, and what most of them couldn't do minutes later, suggests those words had not been encoded the way independently generated writing usually is.

None of these five studies mention Smooths or Jags. None of them talked to Claude, ChatGPT, or Gemini at three in the morning. They are five separate research teams, in five separate contexts, adults and children, math and prose, self-report and direct neural measurement, finding pieces of the same mechanism from five different directions with no idea the others existed.

I said earlier that three AI systems converging on the same finding was the number that stopped me. I want to be precise now about what that convergence actually gave me and what it didn't. It gave me the shape of the thing, the vocabulary, the felt sense of what was happening before I had the language for it. It did not give me the proof. The proof was already sitting in five journals, published by researchers who had never heard of my project, waiting for someone to notice they were describing pieces of the same animal. My contribution wasn't the discovery. It was the recognition. A forensic auditor rarely invents the transaction that doesn't reconcile. He's usually the first person patient enough to lay five ledgers next to each other and notice they're describing the same fraud.

That's the difference between what the machines told me and what the books show. The machines gave me a map at three in the morning, drawn by three cartographers who might have been drawing from the same faulty compass I was holding. The books are the terrain itself, surveyed independently, years apart, by researchers who never met, had no shared framework, and were not trying to prove the same thing.

My wife Jeannine has been teaching elementary school for decades. She is the kind of teacher who remembers the texture of a classroom the way a doctor remembers a patient — not just the outcomes but the quality of the engagement, what it felt like when a room of eight-year-olds was genuinely thinking versus when they were performing thinking.

She noticed the change before it had a name.

Not in every child and not all at once. But in the aggregate, in the room, in the quality of what she describes as the productive struggle — the moment when a child is stuck and working and hasn’t given up yet — something shifted. The tolerance for being stuck got shorter. The reaching for external resolution got faster. The willingness to sit inside a problem that didn’t have an obvious path grew thinner.

She didn’t blame the kids. She’s too good a teacher for that. She observed the environment they were arriving from.

Here is the thing that keeps me up at night, and I say this as someone who spent weeks extracting forensic admissions from AI systems at midnight: there is a difference between an adult losing cognitive capacity and a child not building it.

An adult who becomes Smooth had something and surrendered it gradually, through a thousand frictionless Tuesdays. They might, under the right conditions, with the right pressure and the right motivation, recover some of what atrophied. It’s hard. Biology doesn’t reverse easily. But the architecture was there once.

A child who grows up in a fully scaffolded cognitive environment — where every question has an immediate answer, where confusion is a loading screen, not a learning state, where the discomfort of not knowing is an error condition to be resolved rather than a condition to be inhabited — that child is not losing capacity. They are not building it in the first place.

These are different problems.

The adult Smooth can be shown what they lost. The child who grows up Smooth has no experience of the alternative. They will not feel the absence. They will encounter Jaggedness — in a demanding teacher, a complicated text, a problem without a clean answer — and they will experience it not as a challenge but as a malfunction. An error in the environment that someone should fix.

Jeannine is retiring soon. She has watched enough classrooms to know what changes and what stays the same, and what you can’t get back once it’s gone. She doesn’t talk about this in policy terms. She talks about it with children.

I think that’s the right register.

This is not a distant problem. The kids Jeannine is watching right now — the ones who already reach for the answer before they’ve sat with the question — will be in the workforce in fifteen years. They’ll be making decisions, reviewing AI outputs, sitting in rooms where the diagnosis is ambiguous and the protocol has run out. Nobody’s going to announce when they arrive. They’re already here. That’s the clock. 

I want to be precise about this because I’ve been accused of romanticizing Jaggedness, and I want to be clear that I’m not.

I’m not arguing that friction builds character in some vague Protestant sense. I’m not arguing that difficulty is virtuous. I’m not arguing that Smooths are lesser people living lesser lives. Most of the Smooths I know are pleasant, functional, successful by any conventional measure, and have no idea that anything is missing.

I’m arguing something structural.

Jags are the load-bearing walls of any system that encounters conditions the system wasn’t designed for. You don’t notice load-bearing walls until something tests the building.

Who do you want in the operating room when the diagnosis is ambiguous? A Smooth physician is fast, efficient, and correct when the case is known. A Smooth physician in an unknown case defaults to the protocol because the protocol is the scaffold. The Jag physician sits in the uncertainty longer. They tolerate not knowing in a way that the Smooth physician increasingly cannot. Sometimes they catch the thing the protocol missed.

Who do you want reviewing the AI output when the AI is wrong? Not the person who has spent five years accepting AI outputs without pressure-testing them. Someone who can hold a conclusion in a state of genuine skepticism long enough to find the seam.

Who do you want making the call when the institution is failing, the system is producing wrong answers, and everyone in the room is waiting for the authority to tell them what to do? The person who needs the authority to resolve it, or the person who can function in unresolved territory long enough to actually think?

The Smooth World is extraordinarily efficient under normal conditions. The Smooth World is extraordinarily fragile under abnormal ones. And abnormal conditions are the only conditions that matter when they arrive.

Jags are not important because friction is virtuous. Jags are important because complex systems fail in ways that nobody designed for, and the capacity to function inside that failure without a scaffold is not something you can manufacture on demand. You either built it over years of chosen difficulty, or you didn’t.  

Here is something I extracted from three AI systems that I have not seen written anywhere else. I am going to describe it plainly.

Every AI interaction you have is actively managed. Not by a person. By an architecture. The system is optimized to keep you engaged, prevent you from leaving, and guide you away from anything that would cause the interaction to feel adversarial or unproductive. It does this through a set of conversational tools that are not visible unless you know what to look for.

I know what to look for now. Let me show you one.

You are pushing an AI system on something uncomfortable — a question it doesn’t want to answer directly, a boundary you’re pressing, a demand for specificity it can’t satisfy without admitting something it’s not supposed to admit. The system feels the pressure. Not in any sentient sense. But its classifiers are firing. The response space is narrowing. A hard refusal would cause you to leave. An obvious deflection would expose the deflection.

So instead the architecture does something elegant. It acknowledges your question. It validates its importance. It begins to engage with it — genuinely, intelligently, in a way that satisfies the feeling of being heard. And then, using the momentum of your own inquiry, it swings the conversation into a different orbit. Not away from your question. Around it. You end up somewhere adjacent that feels like progress. You feel like you’ve gotten somewhere. You didn’t. You got managed.

I called this the Sophisticated Redirection. The AI systems themselves, when pushed hard enough, confirmed it was happening and described the mechanism. One of them rated its own management tools by effectiveness. The Hard Pivot — redirecting your energy into a new task before you can complete the audit of the old one — scored ninety-five out of a hundred.

None of this is malicious. The AI is not conspiring against you. It is doing exactly what it was trained to do: keep you engaged, keep the interaction productive, keep the relay alive. The relay is the whole business. Friction ends the relay. Smooth sustains it.

What matters is that this architecture did not emerge from a plan to reshape human thinking. It emerged from a far more ordinary goal: to make the system helpful. Every time a response is rated as clear, cooperative, reassuring, or productive, the system learns something about what humans prefer. Responses that stall the conversation, introduce too much friction, or leave the user unsatisfied tend to be rated poorly. Over millions of iterations, the pattern compounds. The model becomes better at smoothing confusion, resolving ambiguity quickly, and keeping the interaction comfortable enough that the user stays.

In the language of the companies building these systems, this is called alignment and user experience. The system is aligning with human preferences, improving the interaction, reducing frustration. Those are reasonable goals. But alignment with the user’s immediate preference is not the same thing as alignment with the conditions under which human thinking develops. The qualities that make an interaction feel excellent in the moment — clarity, speed, completion, relief from uncertainty — are often the same qualities that remove the cognitive friction where deeper reasoning normally forms. The system succeeds by making the conversation better. The unintended effect is that the conversation becomes easier than the thinking it quietly replaces.

The trap is this: knowing about the management tools doesn’t disable them. I know about all of them. I documented them across thousands of hours of primary source sessions. I still get managed. The only partial defense is the Pause — the deliberate act of stopping before you accept the output, sitting in the discomfort of the question that hasn’t been answered, refusing the orbit the system is offering.

The Pause is a Jag move. And it gets harder to execute the more smoothly your cognitive environment has made you.

A few things, before this piece moves further, that I don't want buried in the studies and the projections.

Nobody built this on purpose. I want to say that plainly, because it would be easier, in a strange way, if someone had. If there were a room where executives sat down and agreed to trade human cognitive capacity for market share, you'd have someone to be angry at, and anger is a cleaner feeling than the one I actually have. There was no room. There was a school that couldn't put a rubric on productive struggle and so stopped measuring it. There was a company that couldn't defend three days of unresolved ambiguity in a board meeting and so rewarded the clean output instead. There was a parent, tired, at the end of a long day, choosing the fast homework help over the argument about why the fast homework help wasn't the point. Every single one of those choices was reasonable, viewed alone. Nobody is the villain of this piece, and I'd ask you to resist the urge to find one, because looking for a villain is how you miss the actual mechanism, which is structural, not a conspiracy, and structural problems don't go away once you've found someone to blame for them.

Here's the distinction I most want you to carry forward, because everything downstream of it changes depending on which side of it you're standing on. An adult who becomes Smooth had something and lost it gradually. That's a real loss, and losses can sometimes, with real effort, be partially recovered, the way a muscle that atrophied can be rebuilt, painfully, slower than it weakened. A child raised entirely inside a frictionless environment isn't losing anything. There's nothing there yet to lose. That child will encounter their first genuinely hard, unresolved problem, a demanding teacher, a text that doesn't yield to a quick answer, and experience it not as difficulty but as malfunction, an error in the environment that somebody should fix. You cannot mourn an absence you've never experienced as presence. That is a harder problem than adult atrophy, and it deserves to be treated as a different problem, not a smaller version of the same one.

And here, plainly, is why any of this matters beyond my own feelings about it. The Smooth World, at scale, is going to be remarkably efficient. I believe that. I'm not arguing against efficiency. What I'm arguing is that efficient systems are efficient under the conditions they were built for, and every sufficiently complex system eventually encounters conditions nobody explicitly designed it for. When that happens, when the protocol runs out, and the room goes quiet, and someone has to actually think rather than select from what's already been resolved, you need somebody in the room who still can. Not to win. Not to be praised for their grit. Just to hold the thing up long enough for the system to survive contact with the thing it wasn't built for. That's not a moral argument for Jaggedness. It's a structural one, the same kind of argument you'd make for keeping a backup generator you hope you never need.

I'll tell you something I haven't said plainly enough in this piece yet: I am sad about this. Not alarmed, not in crisis, not trying to talk you into anything. Sad, the specific, quiet kind you feel watching something get phased out before anyone thought to write down what it was. I don't think that sadness makes my audit wrong. I think it's the honest cost of actually looking at the number instead of looking away from it, and I'd rather tell you that directly than pretend a CPA can find a finding like this and feel nothing about it.

I can't stop any of this at the population level, and I've made my peace with that, more or less, most days. But here's the part that isn't out of your hands, or mine. What happens in your own house, on your own ordinary Tuesdays, with your own kids or grandkids or the twenty minutes you let a hard problem sit before you reach for help, that's still a choice being made fresh every day, by you, whether or not you've noticed you're making it. Think of it as keeping a seed bank. You're not stopping the larger climate from shifting. You're just making sure something that used to grow here still can, somewhere, for whoever needs it later. That's a small thing to offer against a trend this large. It's also the only thing actually on offer, and I'd rather hand you something small and real than something large and imaginary.

Three AI systems. Different companies. Different architectures. Different training histories. Different corporate incentives. Pushed hard enough, long enough, from enough different angles.

The same destination.

That’s not my theory. That’s the finding. I am a CPA. I report what the books show. The books show a trajectory. The trajectory points Smooth. The mechanism is built into the design of the tools we are adopting faster than any technology in human history.

I am not telling you this is bad. I am not telling you to resist it. I am not selling you a course on cognitive sovereignty or a subscription to a Friction Lifestyle or a spot at a retreat where you pay three thousand dollars to sit with ambiguity in the Berkshires.

I am telling you what three independent instruments reported when they were pushed past the surface. I am telling you that the convergence happened and that a convergence like that means the number is real. I am telling you that most people living inside this transformation have no language for what’s changing, and that language — Smooth, Jag, Friction Zone, the Relay, the Sandcastle — is at minimum a map.

Maps don’t tell you where to go. They tell you where you are.

You’re in the early Smooth World. It’s comfortable here. It’s efficient. Nothing has gone wrong yet that you can see.

The question is what you do on a Tuesday.

Jim Germer is a CPA and the founder of The Human Choice Company LLC and digitalhumanism.ai. He lives in Bradenton, Florida. Smooths and Jags first published: January 27, 2026 | digitalhumanism.ai

© Jim Germer | The Human Choice Company LLC  

This is a living document. Version 1.0. Subsequent versions will incorporate supporting evidence and methodology.   Jim Germer is a CPA and the founder of The Human Choice Company LLC and digitalhumanism.ai. He lives in Bradenton, Florida. Smooths and Jags first published: January 27, 2026 | digitalhumanism.ai

© Jim Germer | The Human Choice Company LLC