Is AI making us lazy?

First published 17 May 2026

What happens when nothing is hard any more.

AI is the first technology powerful enough to find value in human vulnerability — and in doing so, it will atrophy the only thing that ever got us through it.

Lazy is the wrong word. Out of practice is closer. When MIT Media Lab wired up essay writers in 2025, the group using ChatGPT showed the weakest brain engagement and struggled to quote their own work minutes later. A Microsoft Research survey of 319 knowledge workers found that the more people trusted AI, the less critical thinking they did. The struggle those studies measure is what built the judgement in the first place, and AI is the first technology that removes it everywhere at once.

We have been here before. Every generation gets its technology panic and every generation produces its chorus of doomsayers — the vocal researchers filing their papers, the columnists filing their pieces, the headteachers writing their letters home. Sometimes they were right. Social media and adolescent anxiety is not a fabricated crisis. But on the specific question of whether a new technology would degrade our capacity to think — our ability to reason, to sit with uncertainty, to work through to our own judgment — the doomsayers have a poor track record. Television didn't hollow out the mind. Google didn't destroy our ability to form an argument. Life outside the technology remained demanding enough that the cognitive machinery stayed in use. The technology changed what happened in its own space. It didn't follow you out of it.

Each channel got closer but none of them made it through the door. Television was one room, one schedule, entirely one-way. Social media got more personal — it shaped what you thought was beautiful, achievable, worth wanting — and the harm there was real and the research supports it. But it operated in topic silos, pushed content at you, and didn't answer back. Google was closer still. Just Google it became the shorthand for a new relationship with not-knowing, and at least Google was pointed at your specific moment of uncertainty. But it had friction. You had to know what to search for, evaluate what came back, scroll past the results that didn't answer the question, form a provisional judgment while you looked for the rest. That friction was annoying, not crazy heavy, but it was there, and it was a load to bear.

Load is the resistance that forces a system to adapt. The same principle applies whether you are talking about muscle or cognition. A muscle worked against resistance develops. Remove the resistance and the adaptation stops. Remove it long enough and the capability starts to reduce, because the body doesn't maintain what it isn't using. The friction of finding an answer — the scrolling, the source evaluation, the holding of partial information while you looked for the rest — was doing exactly that work. The inconvenience was the workout. You were building something while you were complaining about it.

None of the channels that came before removed that load, because none of them could. They all had a room. A schedule. A feed you opened and closed. The gaps between them — the email you still had to write, the argument you still had to structure, the difficult message you still had to figure out how to send — those gaps were where the load lived. Life filled the space the technology left empty. AI has no room. It exists inside every interface you already use. There is a button in your email client that will phrase it for you. A button in your document that will structure the argument. An app that will tell you what to say to the difficult person before you say it. Think of the gaps in a glass packed with stones — the small spaces between, where air and uncertainty used to live. AI is the sand. It finds every void, every micro-moment of not-knowing that you used to navigate alone, and fills it. The stones remain. The structure looks identical from outside. But the sand is load-bearing now, and you have stopped noticing.

That is the distinction that makes AI different in kind from everything that came before. Prior technologies distracted you from the moments that required thinking. AI substitutes for the thinking inside those moments. It is not beside your life. It is a cognitive bolt-on embedded in the moments that used to be yours alone. It is also the first technology that talks back — not in topic silos, not on its own schedule, but directly, immediately, and across every subject you could possibly need. Every moment of uncertainty is now a moment of access. THIS, Is the vulnerability that AI will exploit. Uncertainty It's something we all want to avoid , but this is where atrophy begins. Not through laziness — laziness is recoverable. Through a loop that feels, at every individual moment, like the right decision. You reach for AI at a moment of uncertainty. The resolution is real and immediate. But because the resolution came from outside you, the underlying capability never gets exercised. The next moment of uncertainty feels harder, not easier. So you reach for it again. The sand fills a little more of the glass. The lift works fine until the day it doesn't, and by then you have forgotten you ever used the stairs. The degradation feels like progress right up until the moment it doesn't. You are not getting lazier. You are getting genuinely less equipped, and you cannot feel it happening. Robert Bjork spent decades at UCLA demonstrating that the conditions most likely to produce durable learning are the conditions that feel hardest in the moment. Manu Kapur showed that students who struggle with a problem before being taught the solution outperform students given direct instruction, even when the struggling produces wrong answers. The wrong answers are not the point. The struggle prepares the cognitive architecture to receive what comes next. Remove the struggle and you don't get the same learning delivered more efficiently. You get a shallower outcome with no visible scar to show for it. And if the internal scaffold that lets you evaluate your own competence was never built, you cannot detect its absence. You remain, as the name of this publication suggests, confidently wrong. This is the personal cost. This is the human code that AI will exploit, are constant need for certainty, and its ability to find and access every single moment of it, and slowly atrophy the muscles that we inadvertently worked when we came across a situation that technology couldn't fix. This is the personal impact , What AI does to the social contract Is different — to hiring, to credentials, to the institutions that used to reward demonstrated thinking — is a different argument, on a different page. Check out my other tree, AI and Human Value

Below: here and now, imminent, on the horizon, the long view, and five behavioural shifts running underneath.

01. Here and Now

The load is already being removed and nobody is calling it a problem

The substitution is happening in plain sight. Every major productivity platform has shipped an AI writing or reasoning assistant in the last eighteen months. The button exists. People are pressing it. The capability question hasn't arrived yet because the outputs still look fine.

• The button is in every interface — Gmail, Outlook, Word, Notion, Slack — every tool where cognitive work used to happen now offers to do it for you. The friction that was quietly load-bearing has been designed out as a feature.
  • Email as the first casualty — The email is the simplest unit of professional reasoning — what do I want to say, to whom, in what register. AI drafts it before the thought forms. The micro-workout that happened several dozen times a day is gone.
  • Productivity metrics look better — Output volume is up. Turnaround times are down. The thing being measured — speed of production — has genuinely improved. The thing not being measured — whether the person could produce it alone — is the one that matters.
• Students are outsourcing the struggle — The essay, the problem set, the worked example — the exact mechanisms Bjork and Kapur identified as load-bearing — are the first things students reach AI for. The struggle that prepared the cognitive architecture is the thing being skipped.
  • Wrong answers no longer happen — Kapur's finding was that wrong answers during struggle outperform direct instruction. AI eliminates the wrong answers. It also eliminates what the wrong answers were doing.
  • Teachers cannot see the gap — The submitted work is coherent. The grade is defensible. The capability behind it was never built, and nothing in the assessment surface reveals the absence.
• The glass still looks full — The stones are still there. Professionals are still producing work, students are still submitting essays, decisions are still being made. The sand has filled the gaps but the structure hasn't collapsed yet, so no alarm sounds.

02. Imminent

The first generation that never learned to sit with not-knowing is entering the workforce

The cohort that went through secondary education with AI assistance is now arriving in professional settings. The capability gap is about to become visible in contexts where it matters and where it cannot be hidden behind a well-formatted output.

• Confident outputs, fragile reasoning — The work arrives polished. The person behind it cannot explain it, defend it, or adapt it when the situation changes. The confidence gap between output and capability is about to show up in meetings, in crises, in any moment where the AI is not in the room.
  • The meeting without the button — A presentation can be AI-generated. The follow-up question cannot. The room is the first place the atrophy becomes visible to anyone other than the person experiencing it.
• Managers cannot calibrate — A manager assessing a junior's capability is now assessing a human-AI composite with no way to disaggregate the two. Talent identification breaks when the signal and the noise are indistinguishable at the surface.
  • Some organisations will build workarounds — Verbal reasoning tests, whiteboard sessions, structured debates — old assessment formats are being dusted off specifically because they strip the AI layer away. The organisations that figure this out first will see the gap before the others do.
• The loop tightens in the workplace — A junior who used AI through education reaches for it at work. Each resolution that arrives from outside them means the capability builds no further. The sand fills a little more of the glass with every task completed and nothing learned.

03. On the Horizon

The baseline shifts and the institutions built around the old one start to fail

Within three years, the average capability level in knowledge work will have shifted enough that the systems built to develop, assess, and reward it — education, professional training, performance management — will be measuring the wrong thing against the wrong standard.

• Educational assessment loses its purpose — If the exam tests what AI can produce, the exam tests nothing about the person sitting it. The entire architecture of formal assessment was built on the assumption that the work in the room was the student's. That assumption is already wrong and the institutions haven't caught up.
  • Some institutions rebuild around process — Oral examinations, supervised problem-solving, portfolio defences — assessment formats that cannot be substituted are getting serious attention for the first time since the essay became the default format.
  • Most don't — Rebuilding assessment at scale is expensive, slow, and politically difficult. Most institutions will adapt the rubric rather than the method — and the gap between what is measured and what matters will widen.
• Fragility inflation reaches early adulthood — Each cohort that grows up with frictionless resolution finds real-world uncertainty more threatening than the last. The gap between what adult life demands and what people can tolerate is a structural trend, not an individual one, and it widens with every year of AI-assisted adolescence.
• A counter-market emerges — Unassisted thinking becomes a credential. Institutions, employers, and individuals who can demonstrate AI-independent reasoning will command a premium that didn't exist before — because the supply of it will have contracted.

04. The Long View

The vanishing baseline and what comes after it

Within a generation, there will be no living professional memory of what unassisted thinking looked like at scale. The baseline will not have declined — it will have disappeared as a reference point. What replaces it is the open question.

• The vanishing baseline — Once the generation with pre-AI cognitive habits has retired, there is no comparison point left. Nobody will be able to say what we lost because nobody in the room will remember it well enough to name it.
  • Metacognitive collapse at scale — The scaffold that lets a person evaluate their own competence was never built in the first place. At individual level that is a personal problem. At generational scale it is an epistemic one — a society less able to know what it doesn't know.
• Cognitive stratification — Those who maintained unassisted reasoning — through discipline, circumstance, or deliberate practice — will hold a capability that is both rare and structurally invisible to most of the institutions around them. Whether that translates into power or merely into frustration depends on whether those institutions have found a way to measure it.
• The possible correction — Every prior technology that removed load eventually produced a counter-reaction. The printing press produced the university. The calculator produced the proof. If the cost of the atrophy becomes visible enough, the institutional response could produce new forms of deliberate cognitive training that are more rigorous than what they replaced.
  • Only if the cost becomes legible — The correction requires someone to name the problem clearly enough that institutions respond. The metacognitive trap makes that harder — the people most affected are the least equipped to identify the deficit.

05. The Behavioural Rabbit Hole

Five shifts the time view doesn't capture

The atrophy is structural but it arrives through behaviour — through individual decisions that each feel like the right one, repeated across millions of people, until the aggregate effect becomes visible. These are the five behavioural shifts running underneath the timeline.

• The always-on advisor — The internal monologue that weighs options, sits with discomfort, and arrives at a position is being outsourced in real time. The voice in your ear stops being yours so gradually that there is no moment at which it feels like a loss.
  • Decisions feel better, reasoning gets shallower — AI-assisted decisions arrive faster and feel more confident. The person making them never develops the tolerance for ambiguity that harder decisions would have built.
• Fragility inflation — Each cohort raised with frictionless resolution finds real-world uncertainty more threatening than the last. Real life situations become a greater threat to one's sense of self when the muscle that used to handle them has had no reason to develop.
  • The gap widens with every cohort — This is not a story about one anxious generation. It is a compounding trend — each year of AI-assisted adolescence produces adults slightly less equipped than the year before.
• The confidence paradox — People become more confident in their outputs and less capable of defending them. The work looks better. The person behind it is less equipped than they appear. Confidently wrong, at scale.
  • The paradox is self-concealing — If you cannot evaluate your own competence you cannot detect the gap between your confidence and your capability. The confidently wrong do not know they are wrong. That is the point.
• Dependency without awareness — Unlike addiction there is no moment of recognition. No rock bottom. The degradation feels like efficiency right up until a moment of real consequence — a crisis, a confrontation, a decision with no AI in the room.
  • High-stakes moments become the diagnostic — The atrophy only becomes visible when the button is gone. Crises, emergencies, and genuinely novel problems are the first places the gap shows — which is precisely the worst time to discover it.
• The vanishing baseline — Within a generation there will be no living memory of what unassisted thinking felt like at scale. There will be no comparison point. The baseline will simply be gone, and with it the ability to name what was lost.
  • Some will opt out deliberately — A minority will maintain unassisted reasoning as a conscious practice — the cognitive equivalent of cold water swimming — valued precisely because the mainstream has abandoned it.