Product Friction and the Comfort Crisis
Removing friction is a reliable product instinct, but it can quietly harden into an end in itself. What the psychology of effort, and a book about voluntary discomfort, say about when frictionless is the wrong target.
I read most of Michael Easter’s The Comfort Crisis on the sofa, under a blanket, the heating on and a coffee going cold beside me, with a podcast queued for when I got bored. The book opens with Easter on a windy tarmac above the Arctic Circle, about to be dropped by bush plane into hundreds of miles of Alaskan backcountry for thirty-three days. Grizzlies, glacial rivers, no trail out. I was reading it while optimising my own discomfort down to zero, which is, more or less, his entire point.
The book’s argument is that we have engineered hardship out of modern life so thoroughly that we are now suffering from its absence. We are, in Easter’s words:
Sheltered, sterile, temperature-controlled, overfed, underchallenged, safety-netted.
He builds the case by going to find the people who have deliberately put the difficulty back: the NBA’s top sports scientist using an ancient practice to forge championship athletes, an Oxford economist studying happiness through death in Bhutan, a neuroscientist measuring what extended time in the wild does to a stressed brain, and the bow-hunters who take him into the Arctic. The throughline is a body of research suggesting that the things we have spent centuries removing, hunger, cold, physical strain, boredom, sustained effort, were not only burdens. They were also inputs our minds and bodies are tuned to expect, and their total absence shows up as the modern epidemics of anxiety, restlessness and poor health.
His prescription is not suffering for its own sake. It is to reintroduce discomfort on purpose, in chosen and bounded doses: carry a heavy pack, sit with boredom instead of reaching for a screen, take on something hard enough that you might fail. Every advance that removes a small struggle feels like progress, because individually each one is. Warmer house, faster food, smoother everything. But the sum of all that smoothing is a life with no edges, and Easter marshals a fair amount of evidence that humans do badly without edges.
I am a product manager, not an adventurer, so the thought that kept interrupting my reading was not about the Arctic. It was that removing small struggles, one at a time, each removal obviously good on its own terms, is a precise description of what we do to software. For nearly every product team I have worked with, frictionless has quietly become an end in itself. For most of that history the goal stayed safely out of reach: there was always a step that could not yet be automated, a judgement only a person could make, so “reduce the friction” had a natural stopping point. AI is dissolving that limit. For the first time it is not impossible to picture a journey with no friction left in it at all, which makes a question we could once leave unanswered suddenly pressing: which friction were we ever right to remove? Strip enough of it out of a product, I have come to think, and you can end up with something similar to what comfort has done to us.
The reflex we have stopped questioning
Reducing friction is one of the most reliable instincts in product work. It is not the whole job, or even close to it; a good product manager is usually chasing bigger questions first, whether the thing is worth building at all, whether anyone actually wants it, whether it moves the outcome you set out to move. Nor is the experience itself the product manager’s to draw: shaping the UX is the designer’s domain. What lands on the product manager’s desk is the prioritisation, the call on which problems are worth fixing and in what order, and smoothing a friction point is one of the most common items on that list. That is where the instinct bites. Fewer steps in the funnel. Fewer fields in the form. Fewer decisions for the user to make. Most of the time it is the right call, and I want to be clear about that before I argue with it, because the argument only works if you grant that the default is usually sound. A checkout that takes one tap instead of five really is almost always better. A signup that does not demand a fax number really is better. The history of good product work is, to a first approximation, the history of optimisation and refinement.
The trouble starts when a useful default hardens into an end in itself. Friction stops being one variable you tune against an outcome and becomes a thing that is simply bad, to be driven towards zero wherever it is found. Once that happens you stop asking the only question that matters, which is what a given piece of friction is doing. Some friction is waste: it costs the user effort and returns nothing. Some friction is load-bearing: take it away and the structure it was holding up comes down with it. Treating all of it as waste is how you saw through a structural beam because it was in the way. To be clear, none of what follows is an argument for making products harder, or for designing in struggle so that users feel they have earned the outcome. It is an argument against treating frictionlessness as a goal in its own right, when it is only ever a means, and one that sometimes works against the very thing you wanted.
Problem creep, and the satisfaction you can never reach
A really interesting idea I took from the book is one Easter borrows from the Harvard psychologist David Levari. In a series of studies, Levari and Daniel Gilbert had people pick the threatening faces out of a long sequence, then quietly made threatening faces rarer. People did not relax. They started reading neutral faces as threatening. Their assignment was to identify threatening faces, so they did. The same held when the task was judging research proposals as ethical or unethical: make the clearly unethical ones scarce and ambiguous ones start getting flagged. As a problem grows rarer, we expand our definition of it, so the number of problems we perceive never really falls. The unsettling implication is the one that matters here: we will report that something is wrong whether or not anything actually is. Levari called this prevalence-induced concept change; Easter calls it problem creep.
This is the most important thing a product manager could know about chasing a frictionless experience, and almost nobody frames it this way. Removing friction does not raise satisfaction, because the baseline resets to meet it. Cut the signup from eight fields to three and users will not bank the goodwill; they will start chafing at the email confirmation step that remains, and when that goes, the one onboarding screen left will feel like one too many. Satisfaction was never a function of how much friction there is, only of the gap between expectation and experience, and smoothing just teaches people to expect more. None of which is an argument against improving: continuous refinement is the work, and a substandard experience should never be left to stand. It is an argument against believing that each thing you smooth buys satisfaction, when often it only resets the baseline.
There is a deeper reason this shows up so reliably, and it is the same one behind the comfort crisis itself: deliberately introducing difficulty is not a natural instinct, but removing it is. We see an obstacle and want it gone, rarely pausing to ask what it was doing. That reflex is how we engineered the hardship out of modern life, and it is what surfaces when users tell us what is wrong with a product. They will reliably flag friction as a fault; it is vanishingly rare for anyone to point at a piece of friction and say it is making the experience, or the longer-term outcome, better, even when it is.
This is where a good product manager’s rigour earns its keep, because the task problem creep leaves us with is exactly this: telling the real problems from the ones we are only hearing about because the threshold has moved. The study tells us the defect reports will keep coming whether or not anything is wrong; the judgement is in working out which is which. The discipline is to keep measuring against the outcome you set out to move and to refuse to be governed by individual incidents, whether a loud complaint, a counted click, or a stakeholder who disliked a screen. That is harder than it sounds, because it cuts against how our minds work: we react to the thing directly in front of us rather than holding a line on an outcome agreed weeks earlier. Treating each reported friction as a claim to be tested against the outcome, rather than a fault to be fixed on reflex, is a large part of the job and one of the least natural things it asks of you.
The value was in the effort
There is a second reason effort-free experiences disappoint, and it runs deeper than recalibration: for a great many activities, the effort was the part that made them worth doing. Mihaly Csikszentmihalyi spent decades studying the moments people report as most rewarding, the states of complete absorption he called flow, and they are never the easy ones. The best moments, he wrote, “usually occur when a person’s body or mind is stretched to its limits in a voluntary effort to accomplish something difficult and worthwhile.” Flow arrives when a challenge is matched to your skill and pitched just past your current reach: demanding enough to take all of your attention, not so demanding that you flounder. Too little challenge and attention drifts into boredom; too much and it tips into anxiety or frustration. The reward lives in the narrow band between, in the stretch itself.
That is an awkward finding for a discipline built on making things easier, because it means enjoyment and effort are not opposites. In the activities people care about most, the effort is the enjoyment. A climber in a flow state is not gritting through the ascent to reach a rest at the summit; the absorption on the wall is the point. Take the difficulty out of something that ran on flow and you do not raise the person to a higher state of ease. You remove the conditions that made it worth doing, and what is left is not contentment but a disappointing flatness. The lesson is not that hard is good and easy is bad, but that some of the effort we are quick to design away was carrying the experience, and smoothing it to nothing can drain out exactly what people came for. A product that automates the interesting part of a task and leaves the user the dull remainder has made their day easier and emptier at the same time.
This is exactly where AI-assisted creation gets uncomfortable. When a model does the whole job, the person who prompted it is left holding an output they did not struggle for, and the satisfaction of having made something does not arrive, because they did not make it. There is a related and well-observed problem: the more heavily AI-generated something is, the more generic it tends to be, since the models are trained on broadly the same material and regress towards the same middle. Jake Knapp and John Zeratsky have made the point that differentiation is the strongest predictor of whether a product survives, and that building fast with AI tends to pull teams towards the undifferentiated middle. So the frictionless creation tool can deliver a double disappointment: the work feels unearned, and it is not even distinctive. You have removed the effort that made the activity meaningful and the struggle that would have made the result yours. Some of that lost friction was the human staying in the loop, and it was doing more than anyone noticed: it was potentially elevating the work up from the average of everything the model had ever been trained on.
The friction some products keep
Easter’s own prescription, again, is not suffering for its own sake. It is to reintroduce difficulty deliberately and in chosen doses, and his organising example is misogi: a practice, drawn from Japanese tradition and repurposed by a researcher he meets, of taking on something genuinely hard with a real chance of failure. Leave comfort, meet a challenge, struggle, return changed. The lesson for product work is narrower and a good deal less romantic: some of the effort we are quickest to remove is load-bearing, quietly doing a job, and the only way to know which is to look before you cut.
Look, then, at the products that declined to smooth everything away. Duolingo could hand you the translation and let you tap “continue” until a lesson counts as done; instead it drip-feeds material, drills you with repetition, and builds the experience around a daily streak. Peloton could be a quiet bike and instead puts a leaderboard in front of you. A game with no challenge is not relaxing, it is boring, and boredom is the fastest way out of an experience. The point is not to copy these and bolt difficulty onto your own product. It is that a blanket instinct to strip out friction would have wrecked every one of them, because in each case the effort was the thing the user came for. The UX field has a name for this balance. Nielsen Norman Group frame it as friction and flow: two tools to reach for as the journey demands, friction to slow people down where slowing down helps, flow to speed them through where it does not. In that framing friction is a design tool, sometimes there to prevent a careless action, sometimes there because it is carrying the value, not a defect to be removed on sight.
When the friction is ours
So far I have written as though friction is something we do to users. But the same thing is often pointed back at us, the people who build the products, and the most recent account I have read of what that feels like comes from Teresa Torres. She has spent years arguing that the synthesis at the heart of product discovery, the slow work of turning a stack of customer interviews into a structured map of the opportunity space, is cognitively demanding and worth doing properly. It is, in the terms of this piece, load-bearing friction. The effort is where the understanding forms.
She has also run her own interviews through AI, and she is candid about the result:
When I ran my interviews through Claude, it caught opportunities I missed. But I also caught things it missed. The highest-quality synthesis came from combining both perspectives… I still believe there’s real value in doing the synthesis yourself. But I also know that a draft OST you actually refine is better than a perfect process you never get to. This is about raising the floor.
That is not a small concession from someone whose career is built on teaching the manual craft, and it is the agency question resolved sensibly: keep the human in the work, but let the machine raise the floor. The danger she is steering around is the obvious one. It is not that AI does some of the synthesis. It is letting it do all of it, so that the draft is never refined, the judgement is never applied, and the understanding that only forms through the effort never forms at all. Torres is keeping the load-bearing friction and shedding the rest. That is exactly the discrimination the job now requires, and it is the corrective the comfort-crisis argument needs, so it does not curdle into nostalgia for hardship.
Telling load-bearing friction from waste
None of this is a licence to make products annoying. The Arctic is not better than your living room because it is colder; Easter’s case rests on the discomfort being chosen and purposeful. The same condition applies to product friction. The job is not to add hardship. It is to develop judgement about which friction is load-bearing and which is waste, and to stop treating the distinction as if it did not exist.
A few questions help me sort one from the other. What is this friction producing? If the honest answer is nothing, that the user pays and gets only delay, remove it without ceremony; that is exactly the friction the instinct exists to clear away. But if the friction is producing comprehension, or care, or commitment, or a sense of authorship, or a moment where the user actually understands the decision they are making, then it is holding something up, and removing it will quietly drop whatever that was. A confirmation step on an irreversible action is load-bearing. The effort of arranging your own thoughts before a tool acts on them is load-bearing.
The agency dimension is the one I would watch hardest as we wire more AI into products. Aishwarya Naresh Reganti and Kiriti Badam, who have shipped more than fifty enterprise AI products, frame it as a trade: every increment of autonomy you hand to a system is an increment of control the user gives up, and you should only ask for it once the system has earned the trust. Their counsel is to start with the problem rather than the solution, keeping human control high and agency low at first, and to hand over autonomy slowly. That is the same instinct as keeping load-bearing friction. An assistant that quietly does everything has removed the user’s friction and their agency in the same motion, and turned a creator into a passive curator of outputs they cannot vouch for. The smooth version is not obviously the better product. It is just the smoother one, and we have learned to mistake those for the same thing.
This is why the distinction matters more now than it ever has. For as long as removing friction was genuinely hard, our blunt tools kept some of the load-bearing kind in place by accident, because there was always a step we could not yet take out. AI has the potential to remove that safety margin. Once it is possible to strip out all of the friction, keeping the right friction stops being a refinement at the edges and becomes the heart of the job. And much of what is worth keeping is the friction that holds a human in the loop: the effort that forces a judgement, the step that makes someone own a decision, the resistance that keeps the work from dissolving into the mediocre middle that a model, left to its own devices, will always drift towards.
That mistake is the comfort crisis, ported from Easter’s wilderness into our roadmaps. Each removal of friction looks like progress because each one is, locally. The cost only shows up in aggregate: experiences with no edges, users whose satisfaction baseline has crept out of reach, work that feels unearned because it was. The frictionless ideal is worth questioning not because friction is good, but because “remove all of it” is not a strategy. It is the absence of one.