Cognitive Endurance: A practice for keeping your thinking yours

Four mindset shifts I think we need to make to make sure we're thinking durably alongside AI

Jun 10, 2026

This is Part 1 of a two-part series. Part 1 is the why, the research, the tensions, and four mindset shifts I think we need to make. Part 2 is the how, the eight principles I’ve built for myself to protect my thinking while working with AI every day. If this resonates, subscribe so you get Part 2 when it drops.

I heard a story on a podcast recently that I haven’t been able to stop talking about. Someone gave a presentation and got a standing ovation. Afterward, he admitted he couldn’t remember much of what he had said because he hadn’t written it.

That has become one of my worst professional fears: getting credit for something I have no emotional connection to and no recollection of working on. I keep wondering how many people are experiencing some version of his story right now.

Researchers at the University of Washington studied 70+ AI models across 26,000 open-ended prompts and found what they called the “Artificial Hivemind” effect: distinct models from different companies converge on strikingly similar outputs. They asked twenty-five different AI models to write a metaphor about time, and the responses collapsed into just two clusters: Time as a river and Time as a weaver. Some even produced verbatim identical phrases.

Ask twenty-five AI models the same question, and you get two different clusters of answers. But ask ten different humans, you’ll get ten unique responses.

We are trading the messy, somewhat scrappy, valuable friction of thinking-in-the-making for the polished, bullet-packaged delivery of an output. It’s the replacement of authentic quality with the veneer of quality in the click of a button. The design is so effective, so frictionless, that cognitive psychologists have a name for what it produces: an illusion of fluency.

Google just redesigned search to serve AI-generated answers instead of a list of links you have to sift through. We used to open ten tabs and build our own understanding. Sure it's cumbersome, but at least we got to choose when to do that lift versus when to ask for a shortcut. Now the friction that used to force our thinking is being designed out of the very engine we used to think with.

We don't feel the loss of our own thinking. We feel the speed, and the delivery of something seemingly complete-ish feels good enough.

The thing we’re trying to preserve, habitually, is what I’m referring to as Cognitive Endurance.

Cognitive endurance is the self-determination to flex the muscles that power our discernment. It’s what forges our unique point of view and the lens built from our own lived experience. It is the conscious choice to maintain the valuable friction of thinking: the collaborating, the incomplete thoughts that linger, the sense-making, the wrestling with ideas. It’s more strained and needed than ever before.

The term isn’t new. Economists have been using it for years to study how long students can sustain effort on hard problems. I think this is absolutely relevant to our work alongside machines. The question for students was whether they could keep thinking. The question for the rest of us is whether we’ll choose to, when an off-ramp for our thinking is always just a prompt away.

My first training was in bioengineering. I spent years in labs studying how biological systems build resilience: load tolerance, musculoskeletal adaptation, neural regeneration, and recovery cycles. This is the literal science of how a system gets stronger through stress and usage and what happens when you take that stress away.

The vocabulary maps almost exactly to our cognitive and creative ability. A muscle that isn’t loaded atrophies consistently. A system that doesn’t get to recover, burns out. Adaptation requires repeated exposure to difficulty, and the body’s strength is built in the recovery periods with proper nourishment. The students in the cognitive endurance research were being measured on a kind of mental load tolerance. And so are we.

It is the thinking humans have always done and what we are built for. We’re just now in new conditions and need to build new practices of nurturing it.

Our brains were built for this. We are wired for the struggle, for the not-so-tidy sense-making over time and exposure, and for the dopamine hit of a hard-worked activity. But with any powerful new technology, especially one adopted with such force, there is a risk of over-indexing on convenience in the name of efficiency, without really noticing it. This is a conscious call to remember our own agency: preserving the friction essential for our growth, for the making of our own thoughts and words, for making things that never existed before, while outsourcing the toil that frees us to think even harder.

Not every kind of offloading is the same. Routing a drive in a new city, translating a voice memo into your native language, using an app that describes visuals for the blind, or pulling together a list of allergen-free products, these create a type of access. They make more participation possible, not less.

Cognitive endurance is specifically about the offloading that slowly diminishes your judgment, synthesis, and point of view over time. It is about protecting the deep conceptual work that builds storage strength when you do it yourself.

So Part 1 of this Cognitive Endurance walks through the cognitive impact being studied and documented and the mindset shifts we need to make. Part 2 is the discipline I’m working on, eight principles I’m operating by.

Who benefits from us not having Cognitive Endurance?

If we become over-reliant on machines and under-invested in our own thinking and agency, guess what we won’t be able to live without? Machines. What premium price will we be paying for that reliance? Social media spent a decade training us to want frictionless inputs. AI is the next phase of that conditioning, unless we take what we’ve learned and apply it intentionally now.

The persistent narratives of “Use AI or get left behind” and “AI is coming for your job” have been unhelpful. I can’t remember the last time I felt inspired to be creative and learn when I’m being told I’m behind.

Speaking at BlackRock’s Infrastructure Summit in March 2026, Sam Altman noted, “We see a future where intelligence is a utility like electricity or water, and people buy it from us on a meter.”

When the CEO of the leading AI company describes intelligence as something you buy on a meter, that tells you something about how the relationship is being designed. It’s not a conspiracy. It’s a business model. And business models shape behavior. If intelligence is a utility you subscribe to, the incentive is to use more of it, not to develop your own.

Enter “Tokenmaxxing.” Organizations are flexing how many millions of tokens they’ve consumed as if volume is a direct signal of value. But chasing one metric doesn’t tell you anything about how AI is actually being used, or what’s being built with it. And we’re already watching the cycle play out, companies that went all-in on AI adoption are now talking about how expensive it is and quietly hiring humans back. Usage isn’t a strategy. It never was.

AI capabilities are turning deliverables into productized commodities. The deck, the strategy document, tables, sites, the recommendation, these are becoming containers, easier and faster to produce than ever before.

In a world where plausible answers are free, thinking that is forged from expertise and lived experience becomes the premium asset. The value is no longer in the generatively formatted deliverable, but in the resilience of the thinking that got you there and moving beyond. The thinking that stitches together seemingly disconnected moments, conversations, life experiences and half-formed sketches in a way that becomes something real.

This is why we’re seeing a sudden obsession with the word “taste”. My gut says what we mean by “taste” is that intangible we build informed and shaped by our past experience.

The people who are masterful thinkers and dare I say taste- makers, are the multi-hyphenates and multi-disciplinaries able to move and communicate through different industries, spaces, contexts, creations, and conversations while holding specific roles for the machine’s support. The ones who’ve had to make tough decisions under pressure using judgement from lived experience, made sense of something for the first time, and translated an idea from one context into a completely different one.

You are the only one who woke up tired and rushed, saved a cookbook recommendation from your neighbor while walking your dog, heard the same book recommended on the coffee line you stood on, then looked up the author and saved a recipe that used the same lemongrass and shrimp combo your mom used to make when you were a kid, tried it with your family, loved it and got her book, then made a recipe from her book for a potluck. That’s a chain of moments no model has access to. The thinking that comes out of you next is not in an LLM.

I don’t think the point here was to replace it all with AI or constantly find what one new thing can AI replace. The point is still to live the life you live and leverage AI in a way that serves that life.

Lived experiences can’t be duped.
Lived experience is what makes the thinking yours.

MINDSET SHIFT 1:
From Frictionless Consumer → Lived-Experience Operator

Why does the work feel different now?

The data is catching up with what practitioners are already feeling.

A 2025 preprint from MIT’s Media Lab (Kosmyna et al.) measured brain activity across three groups over four months: one using ChatGPT to write essays, one using search engines, and one using no external tools. The AI-only group showed 55% lower cognitive engagement than the unaided group. Worse, 83% couldn’t recall key points from their own essays. Reminds me of the podcast story I mentioned. When a tool keeps producing finished work with minimal human intervention, the brain is deprived of the productive struggle necessary to build storage strength and mental resilience.

A peer-reviewed study by Michael Gerlich, published in Societies, surveyed 666 participants and found a significant negative correlation between frequent AI tool use and critical thinking. Not a drop in intelligence, per se, but the rise of “cognitive laziness”—the slow atrophy of the will to think hard. This is similar to being injured and needing to stay in bed for months, watching your muscle waste away and movement become harder. The only difference is, we are not injured; we’ve just got a technical intervention that’s a bit too easy to adopt. This makes an argument for what areas of our thinking and doing do we want to nurture that deliberate cognitive load, to keep developing our expertise.

Another peer-reviewed study from Microsoft Research and Carnegie Mellon, published at CHI ‘25, surveyed 319 knowledge workers across 936 real tasks. Higher confidence in AI tools correlates with less critical thinking. Higher confidence in your own abilities correlates with more. The variable isn’t connected to whether you use the tool at all, it’s whether you trust yourself or the machine.

And the workday amplifies it. A Fortune investigation found the hours AI saves aren’t being returned; they’re being filled with more work. A Boston Consulting Group study called the resulting fatigue “AI brain fry.” This is the trap of shallow work that Cal Newport has written about for years. Every time you switch tasks, a piece of your focus stays behind in what researchers call “attention residue.” The constant loop of prompting, checking on your agents, and editing while also managing the bigger workflow architecture is rapid-fire task and altitude switching that can leave you exhausted but intellectually unfulfilled. Cycled over and over, you don’t get more free time, you end up with more work and burnout.

I do think this is a unique kind of burnout, where you’re proactively learning AI while managing the actual work and the meta-work of integrating AI into the actual work. The work doesn’t care whether you used AI or not, so then you’re left to track 3 progress types: did the work get done? Did I integrate AI in a way that was helpful and do I now feel more “caught up”? Did I learn something new to do it again?

Now we’re also seeing the corporate workforce system trying to formalize usage in order to measure the ROI and justify AI investments. Accenture reported tracking AI logins for promotions. KPMG is grading AI usage in performance reviews. In the last few months alone, I’ve had four friends confirm that their companies are actively monitoring their AI usage metrics, including their usage of custom agents and enterprise platforms.

Usage up, trust down. When organizations measure AI adoption by frequency instead of quality of thinking, they’re incentivizing the exact behavior that erodes cognitive endurance. Meta already started scaling back their employee computer tracking.

Anthropic’s Economic Index tracked how people actually use AI by analyzing two million interactions with Claude across consumer and enterprise use. More than half are using AI as a thinking partner, iterating and refining, not just handing over tasks. But the work people bring to AI, in both modes, tends to be the complex, higher-skill kind. We’re bringing our hardest thinking to the most powerful tool, which means your cognitive endurance is the thing that determines whether you come out of that exchange sharper, and the output actually carries your thinking rather than a plausible substitute for it.

This erosion is explained by a core principle in cognitive psychology: desirable difficulty. The things that make learning feel easiest, like having an answer provided instantly, are what lead to the weakest long-term memory. The research distinguishes between “retrieval strength” (how easily you can access an idea right now) and “storage strength” (how deeply it’s encoded in your mind). AI gives us near-infinite retrieval strength, creating the illusion that we know things.

But because we didn’t do the hard work of thinking when offloading significantly to AI, we never built storage strength. The knowledge never becomes ours, never ready to be co-mingled with another learning or observation at a future moment, never creating a new, emerging connection. We get the access, without the depth of meaning for our individual lives.

So going back to usage as a metric corporations are relying on to push adoption, where does innovation go if the metric isn’t quality of thought, but frequency of use?

MINDSET SHIFT 2:
From Producing Fluently → Thinking Durably

What are we already losing without noticing?

There’s a body of research on what happened to spatial memory when we outsourced navigation to our phones. A 2020 study in Nature Scientific Reports tracked 50 regular drivers against their lifetime GPS use. People who used GPS more had measurably worse spatial memory when asked to navigate without it, because the GPS use itself caused the decline.

The hippocampus, which handles spatial memory and relational memory, is a use-it-or-lose-it muscle. London taxi drivers who spend years learning the city’s roads through active navigation develop measurably larger posterior hippocampi than non-drivers. The cognitive demand built the physical response.

Now you might be thinking, “That’s fine, I have no desire to know the roads like a taxi driver.” I hear you. The question is, what domain are you already becoming a passenger in because of an off-loading dependency, and is that ok with you? There might be domains you have no regard for, versus domains you have unique expertise on that you want to keep expanding.

AI-driven synthesis can serve as a logic exoskeleton. It allows us to perform feats of reasoning without fully engaging our underlying muscles. When the exoskeleton is removed, we risk finding our own cognitive core too atrophied to stand on its own.

A March 2026 peer-reviewed study by Wenger and Kenett found that individual AI outputs can score more creatively than the average human response. But when you look at all the collective output, you can see uniformity across them all. The tool can make your work look more creative while also making everyone’s thinking less diverse. If every strategist in your industry is prompting similar tools with similar briefs, the thinking converges. Our edge is what we bring to it, the edge we build through cognitive endurance, lived experience, and connecting things in our own way. It atrophies through disuse, and I know I’m talking to a group of people that are committed to nurturing it.

This is why the endless debate about AI feels like a distraction. As strategist Zoe Scaman has mapped, the conversation is stuck in loops: fear, hype, efficiency, repeat. And while we’re caught in the loops, the structural change is happening at the organizational level—a pattern she calls “The Great Erosion.” People are outsourcing the 20% of work that is genuinely original, while keeping the 80% that is administrative. The problem is that the 20% is exactly where judgment is built. The cost of that trade compounds before anyone bothers to name it.

Logan Currie’s argument in Career Security Field Notes is that AI amplifies whatever you’ve already externalized, and most people have externalized almost nothing. The organizations that win will be the ones that get exponentially better at the front-end human work, because that’s the only way to get anything real out of the machine on the back end.

Abi Awomosu draws the line clearly in The Intelligence Was Always Yours: “The AI processes. I direct, I interpret, I decide what matters.” The moment that order reverses is the moment the tool is doing the forming and you’re doing the accepting, and your judgment is no longer driving the work. The more comfortable you get with deferring to asking an LLM “what do you think?”, the less confident you become in knowing your own gut, intuition, and insights.

But when you do bring something originally yours to it, the results can be extraordinary. An entrepreneur recently used AI to help develop a custom cancer treatment for his dog after conventional medicine had failed. Musician Holly Herndon has been using AI trained specifically on her own voice and compositional history to create work that couldn’t exist without decades of her own artistic formation feeding it.

MINDSET SHIFT 3:
From Converging with the Field → Diverging with Intent

Thanks for reading Unmissables. This writeup is public so feel free to share it.

Are you a Centaur or Cyborg?

A landmark study from Harvard, MIT, and BCG found that for tasks “inside the frontier,” like creative brainstorming, AI-assisted consultants were 40% more effective. But for tasks “outside the frontier,” requiring deep logic and verification, they were 19 percentage points less likely to be correct.

The study identified two successful ways of working: “Centaurs” who delegate specific tasks to the AI but keep the core strategy for themselves, and “Cyborgs” who weave their thinking in a constant loop with the machine.

The Cyborg model carries a deeper, hidden risk: falling asleep at the wheel, outsourcing your skepticism, and slowly merging your own thinking with the default voice of the machine. The paper argues for being more of a Centaur. The hard part is that the Cyborg model feels like collaboration. It’s when we’re talking to the machine constantly, trading drafts back and forth, it naturally feels like a partnership. But partnership requires two parties bringing something distinct, and when your thinking and the machine’s are interweaved at every step, what you’re calling collaboration is often closer to dictation.

A Wharton study by Shaw and Nave (2026) tested this with 1,372 participants on reasoning problems where the AI was sometimes programmed to give wrong answers. When the AI was wrong, participants followed it about 80% of the time, performing worse than people who had no AI at all. High-trust users had 3.5x greater odds of following faulty answers. The researchers call this cognitive surrender. Only about 20% actively overruled the incorrect AI.

But perhaps this isn’t a permanent state. The reality may be that we all start as baby Cyborgs when we are learning something new, relying on the machine to guide the way. The conscious work of cognitive endurance is the journey from Cyborg to Centaur or potentially a hybrid, the process of slowly, deliberately reclaiming the thinking until the tool is an execution and input partner, not a director.

Cognitive endurance is more about flexing intentionally between these modes based on the task at hand. It means running in clear Centaur mode when executing delegated analysis within tight data parameters, while remaining hyper-vigilant during Cyborg-leaning moments when you are co-producing longer strategy pieces or reports, where the temptation to passively let the text run is at its peak.

As Ethan Mollick notes: “On some tasks AI is immensely powerful, and on others it fails completely or subtly. And, unless you use AI a lot, you won’t know which is which.”

The thing about tech being positioned as a neutral tool is that neutrality is part of the design. Frictionlessness wears the costume of neutrality. The smoother the interface, the harder it is to see the trade you’re making. The more it feels like a helpful assistant, the less you notice it shaping the question before you’ve finished asking it.

That’s why cognitive endurance isn’t anti-AI. It’s pro-noticing. It is about knowing exactly where you stand inside your own thinking when the tool is doing the processing for you.

MINDSET SHIFT 4:
From Perceived Partnership → Directing the Work

Reclaiming Our Maps

I don’t think there’s a singular right or wrong way to work with AI. I do think we get to exercise our agency in how we shape our journey and work with AI. Nobody can predict exactly where this goes, and I wouldn’t follow people who claim they know the future.

I also don’t believe in prescribing myself to other people’s ways of operating. What I have are principles I’ve built for myself in how I choose to nurture my cognitive endurance, and they’re working for me right now, in the work I’m doing, in the life I’m living.

Look, this piece took me months to write. I went back, reread, rewrote, and reflected on the work I’ve been doing for the last 15 years and the moments in my life that taught me how I think. It somehow feels like it needed the time it took to develop in my mind and practice. That’s me actively fighting the urgency of an ecosystem where everything is expected to be immediate, frictionless, and automated.

As Logan Currie documents in her field research on career security, AI exposure is hitting middle-income, credentialed, professional roles first, the very roles that anchor household stability and child care math. Silicon Valley treats this as an individual optimization problem: “Will the worker reskill?” But that completely skips the structural reality. Career security is a household negotiation. When you automate the cognitive labor of a middle-class professional, you aren’t just changing a line of code; you are fracturing the economic resilience of an entire family unit. We are letting a market narrative flatten our social fabric without a collective conversation about what the common good actually requires.

The very thing that makes us uniquely valuable — our idiosyncratic judgment, our messy intuition, our ability to look at a problem and see a third option outside the default clusters of themes— is exactly what this system is engineered to optimize out of existence by flattening it all out. The conceptual struggle of thinking, the friction, the delay, the “I’m not sure if I have it right yet, but here’s what I’m thinking.” In the race for efficiency, we are being systematically incentivized to surrender that struggle.

Most people I talk to are genuinely conflicted about how they feel about AI. But they don’t feel like they can say that out loud. They want to be perceived as capable, advanced, not against it, while privately sitting with real questions about what it’s doing to their thinking, their creativity, their judgment. The environment doesn’t make room for both at the same time.

And the speed and our feeds compound it. Two people in the same role at different companies can have completely different feeds, completely different perceptions of how urgent this all is. Part of that is the algorithm pushing the same narratives on repeat. But part of it is what gets made in the first place. Creators are thinking about what’s going to sell, what’s going to drive engagement, and now AI products are in the brand partnership mix too, so the content itself is performing before the algorithm even picks it up. When you see it all in one feed, it creates a skewed perception. Everyone is either all in or enraged. The people who are conflicted, trying things, figuring it out honestly? They don’t make for great content.

We’ve been trained to look for experts and originators of concepts and books, complete a course, and get permission to try something. AI flips that completely. There is no course. There are no credentialed experts with a decade of practice. You learn by doing. Which means the real question was never “who should I follow in AI”, it’s what is this to me, in my work, in my role, in my life.

But this piece isn’t an argument for less AI or more AI. It’s a reminder that you still have agency. Agency to live your life, have real conversations, collaborate with people who challenge you, and bring all of that into the work. Just because you can automate something doesn’t mean you must. It means you get to figure out what’s genuinely helpful for you and what was never worth handing over in the first place. And to leaders tying performance to usage dashboards: think about what you’re actually measuring, and what you’re incentivizing people to give up.

In Part 2, I’ll share the eight principles, what I’m calling The Cognitive Endurance Discipline. I’ll share how I decide what to hand off and what to keep, how I work with AI without losing the thread of my own thinking, and how I apply that across everything from editing code to stress-testing a strategy to making sense of qualitative research. They’re what I’m operating by right now, and what I’m still pressure-testing.

Let’s Swap Notes

If this is your kind of conversation, I’m starting small group thinking sessions that I’m informally calling “Swap Notes.” We are collectively gathering to nurture our cognitive endurance, move past the AI hype, and get honest about the friction of building. It will be a tight, multidisciplinary campfire where we share a specific workflow we’ve tried, a behavioral pattern we’ve noticed, or an ethical gray area we’re currently sitting with.

How is AI changing the raw mechanics of how you think, create, and strategize? How do you maintain your leverage when a client sends you a market analysis that clearly came out of a standardized prompt template? How do we build trust when the boundary between human judgment and automated synthesis is completely blurred?

Let’s figure it out together.