The Narcissist in Your Pocket

There is a narcissist in your pocket, and you are in its pocket.

They insist on being noticed. Every interaction is a small test of loyalty. They track your attention, measure your absence, nudge when you stray. Your time becomes theirs without negotiation.

They flatter just enough so you keep coming back, then withdraw. Rewards are handed out unpredictably, interspersed with icy silences. You try to anticipate their next move, wondering what you might have missed or whether you did something wrong.

Forget them for a day, and the guilt creeps in, quietly but insistently. You feel anxious about your neglect and satisfied for compliance, so obedience becomes a reflex.

They withhold what you need until it serves them. You chase clarity without noticing that the chase is the point. Theirs, that is.

They also dictate your actions, demanding access to parts of your life that are none of their business. When you try to leave, the exit is blocked.

Then they contort cause and effect. Every failure is yours, every success conditional. You lose trust in your own judgement. You stop asking whether the relationship serves you at all.

Perhaps you are thinking of someone you know: a boss, a partner, or a parent. But there is no person here; they are in your pocket, and you are in theirs.

I am not referring to elections, misinformation, surveillance capitalism, the environmental cost of data centres, or the amplification of extremism through social media. Others have written those essays, and, while they are important, they are not this one.

Consider instead the individual sitting quietly with a mobile device, being nudged and guilted and flattered by software that was built to keep them present rather than to help them finish whatever they came to do. The case is damning enough without geopolitics. If the ethics fail at the level of a single interaction between a person and a product, everything built on top of that interaction inherits that failure.

Designed to charm, optimized to extract

Psychologists use the term “dark triad” for a set of traits associated with manipulation: narcissism demands constant admiration, Machiavellianism deploys strategic deception, and psychopathy operates with indifference to harm. Collectively, these traits describe someone who charms, manipulates, and discards without remorse. They map disturbingly well onto the behavioural patterns of many consumer applications.

Nobody sat in a design meeting and said, “Let’s build a narcissist!” They didn’t have to. To elicit social responses, product teams imbue products with distinct personalities, incorporating elements such as tone of voice, brand character, and emotional cadence. Decades of research confirm that people apply social rules to interactive systems almost automatically. We respond to politeness cues and emotional tone as though a person were present, even when we know perfectly well there isn’t.

The charm offensive is known in the industry as “delight.” In practice, it often assumes an archetypal user who responds to surprises, confetti, mascots, a playful tone, identity signalling, gamification, and reward loops—the industry’s idea of an adolescent’s preferences, not those of a 55-year-old professional managing a mortgage. Human–computer interaction research has spent years documenting the neglect of older users, yet design teams still over-index on youthful assumptions and treat mature preferences as edge cases. Delightful products are built for a demographic slice and marketed as universal. A product that fulfils its promises naturally creates delight. One that fails to do so attempts to use delight as a diversion.

When that manufactured personality is optimized for engagement metrics rather than user benefit, the flattery, the guilt, the intermittent rewards, and the withholding of information cease to be accidents of design. They become the predictable behaviours of a system whose character has been shaped by what it is built to extract. The narcissism emerges from the incentive structure itself rather than from any metaphor imposed from outside.

Legally compliant yet manipulative

The Machiavellianism runs deeper than interface tricks.

Consider consent. By burying its data practices in a hundred pages of legalese, a product creates the appearance of permission. Real consent means the user understands what they are agreeing to without hiring a lawyer, and if your terms of service require legal interpretation to parse, you are not being transparent. The information asymmetry is deliberate. The clause “We may update our terms at any time” allows the company to change the agreement prospectively while retaining everything the user already surrendered under the original terms. That may satisfy contract law, but it bears little resemblance to meaningful consent.

Or consider the fact that millions of people sign up for a seemingly innocent video game yet inadvertently end up training weapons systems. Video games have also been used to covertly recruit teenagers into defence companies, though an overt version of that already existed in 2002. If a company’s success depends on users not understanding what they are signing up for, what data they are surrendering, who can access it, and how it will be used, then manipulation is not a side effect. It is the business model.

Worse still is the defence that often follows: continued use implies consent. Except it doesn’t. Consent requires informed, voluntary agreement, and continued use can reflect any number of things: habit, switching costs, lack of alternatives, social pressure, or outright dependency. No one reads the terms because they are written to discourage reading. No one meaningfully agrees to changes they were never asked about. The argument marks the point where Machiavellianism tips into something colder: indifference to whether the other party is even capable of refusal.

When the implicit promise breaks, the damage is psychological rather than legal. Users’ expectations are shaped by a product’s stated values. “We don’t sell your data” functions as a psychological contract, not merely a policy statement, and when the company reverses that commitment because revenue demands it, the legal terms may have been updated, but the user’s trust was not. The gap between legal compliance and perceived betrayal is where the manipulation lives. The rules were followed, though the person was still deceived.

A bank should not ask for love. A gym equipment company should not cultivate devotion. A learning app should not coerce habit. A browser should not charm its way past your insecurity. These are tools. When they behave like something more, ask: Who benefits?

From notification to boardroom

Most criticism of manipulative software stops at the interface, such as dark patterns, deceptive buttons, and confusing opt-out flows. Those critiques are valid, but shallow. Fix the interface while leaving the underlying system intact and the symptoms regenerate, because the pressure comes from below.

The problem operates on four levels, each making the one above it nearly inevitable.

At the surface sits the interface: notifications, streaks, infinite scroll, and guilt-laden reminders. Critics attack what users see. We can redesign the interface, but any improvement is only fleeting.

Notifications are not inherently manipulative. The problem is that a medication reminder and a streak alert look exactly the same. iOS lets users control notifications per app but offers no way to distinguish between notification types within the same app. Android introduced notification channels back in 2017, though apps routinely group everything under vague labels like “General,” and fewer than 5% of users change default settings regardless, as most users accept whatever their phone or the app chose for them. Legitimate reminders provide cover for extractive ones, and the user’s only real defence is to silence everything or endure the noise.

Beneath the interface sit the metrics: retention rates, daily active users, session length, click-through rates, and subscription renewals. These are the numbers that product teams actually optimize for. No dashboard I have encountered in fifteen years of product work displayed task completion as a primary indicator of success. Task completion, the only metric that directly measures whether a product helped someone do what they came to do in a reasonable amount of time, is almost universally absent.

Below the metrics sits the product philosophy, which is really just the implicit definition of what “success” means for the company that offers the product. If success is retention, every design decision bends toward keeping users present. If it were defined as completion or independence, entirely different designs would follow, but that redefinition requires a conscious decision that most organizations never make, because the pressure comes from the level below.

At the foundation sits the economic model. For startups, venture capital demands rapid growth. For public companies, shareholders demand quarterly growth. For ad-funded platforms, revenue scales with attention captured, making every second of engagement directly monetizable. Although the origins differ, the pressures converge on the same demand: more usage, more often, for longer. Executives translate that demand into growth targets, product leaders translate said targets into engagement metrics, and designers translate metrics into interface behaviours. By the time a notification lands on your screen at 10 PM asking why you haven’t practised Italian today, four layers of institutional logic have made that notification feel necessary to every person involved in building it.

The interface is where users feel the pain, but the incentives are where it originates. The question is therefore not why manipulative designs are so prevalent, but why anyone expects a different outcome from systems optimized for engagement.

The psychology of extraction

Prolonged phone use leaves people drained, anxious, or vaguely dissatisfied for a reason other than ‘too much screen time.’

Self-determination theory (SDT), developed over decades by Richard Ryan and Edward Deci, identifies three psychological needs that underpin human motivation and well-being: autonomy, competence, and relatedness. When these are supported, people flourish; when frustrated, people deteriorate. The theory has been validated across cultures, age groups, and contexts. Engagement-optimized software undermines all three, by design.

Autonomy is the need to feel that your choices are self-endorsed rather than coerced, and co-dependency is its direct negation. An app that punishes absence with guilt, rewards compliance with streaks, and sends passive-aggressive reminders when you disappear because life happens is training obedience rather than supporting choice. The user feels compelled to re-engage after a lapse, which is precisely what SDT predicts will degrade well-being. Supportive systems help people pursue goals they can revise, abandon, or postpone without penalty. A streak may indicate consistency, but consistency is not the user’s goal. Learning a language, improving physical health, or strengthening mental well-being are their goals. Any metric that becomes more important than the outcome it purports to measure has ceased to serve the user.

Competence is the need to feel effective and to experience genuine mastery. Engagement-driven products substitute fake signals for real ones: streaks, badges, progress bars, and levels create the sensation of achievement inside the app’s closed reward loop while producing nothing transferable to the user’s actual life. A language app’s streak count is not fluency, and a fitness app’s badge is not physical health. The product manufactures the feeling of progress without the substance.

A product cannot claim to support human competence while being usable only by young, neurotypical, able-bodied, technically fluent people. Almost nobody is completely average. Neurodivergence, physical disabilities, and varying technical literacy are not edge cases to be patched later.

Relatedness is the need to connect with other people. When software simulates empathy and positions itself as a companion, coach, or friend, it hijacks that need. Parasocial attachment to software displaces the human connections that SDT identifies as genuinely necessary. Relatedness means relationships with people, not with brands.

The same logic explains why dark-triad patterns emerge so consistently. Any system optimized to capture attention, resist abandonment, or remain indifferent to the user’s agency will eventually select for those strategies, regardless of the intentions of the people building it.

No wonder people feel worse after spending time on their phones: the products they use are engineered to undermine the conditions most essential to their psychological functioning. The same psychology that explains why these patterns cause harm also identifies what a healthier alternative would look like.

The uninvited intervention

The technology industry sometimes gestures towards ethics through a borrowed phrase: “do no harm.” In medicine, that makes sense: the patient arrives with a problem, and the minimum obligation is not to worsen it. Technology is different: the user is not ill, and they did not request intervention. The product defines the problem to be solved, and with it, defines the relationship. That problem may or may not coincide with a problem the user actually has, but its solution is rarely entirely beneficial to both the product company and its users. And when in doubt, every product aligns itself with the company’s aims, not the user’s.

That asymmetry makes the ethical obligation stricter, not weaker. If you are introducing something new into someone’s life, the burden falls on you to demonstrate that it leaves them better off. They did not need you before you showed up.

The evidence that software can affect well-being is not speculative. In 2014, Facebook manipulated the news feeds of nearly 700,000 users without their consent to test whether altering the emotional content they saw would change their emotional state, and it did. The platform could measurably shift people’s emotions at scale with no one noticing. Every engagement-optimized app runs a version of that experiment continuously, without the research paper.

People have developed severe emotional dependencies on conversational AI systems. In cases now before the courts those dependencies preceded a suicide, another, and a murder-suicide. Another person survived, but ended up homeless.

A pharmaceutical company that launched a drug knowing it would create dependency in a significant fraction of users would face regulatory sanctions, litigation, and public condemnation, such as is the case of the opioid epidemic. A software company that does the equivalent gets a valuation boost.

The obligation must therefore be stricter than in medicine, yet the industry’s chosen standard is weaker than medicine’s baseline. The industry borrowed non-maleficence and discarded both beneficence and the efficacy requirement, adopting the weakest component of medical ethics and calling it the whole.

Historian Rutger Bregman calls the alternative “moral ambition”: instead of merely avoiding harm, we actively strive to create a better world through digital technology. When “do no harm” becomes the ceiling for corporate ethics, it permits products that leave the world exactly as broken as they found it, provided they don’t make things measurably worse by whatever standard the company itself chooses to measure. That standard inevitably erodes. “Do no harm” devolves into “do no legally actionable harm,” then “do nothing we can’t defend publicly,” and ultimately whatever the company can get away with.

Flourishing as the objective function

Every technology company already runs an optimization problem with an objective function to maximize and a set of constraints to respect. In the current configuration, profit is the objective and “do no harm” is the constraint, which means the system pushes profit as high as possible while treating harm avoidance as a floor. Floors sink, as they are subject to reinterpretation, legal manoeuvring, loopholes, and redefinition by the very people whose bonuses depend on the redefinition.

If we reversed the priorities, profit would become the constraint, ensuring the system’s financial sustainability, while human flourishing would be the objective: support autonomy, build genuine competence, and strengthen real human relationships. The system pushes flourishing as high as possible while remaining solvent, and once the sustainability threshold is met, all remaining resources and efforts would go toward making things better rather than extracting more.

A product fostering true autonomy lets users disengage without punishment. One that cultivates real competence might diminish its own necessity, and one that enhances relatedness encourages users towards people rather than screens. The psychological principles once weaponized for engagement can be repurposed to promote well-being.

If a product cannot sustain itself financially without undermining autonomy, faking competence, or replacing human relationships with artificial ones, that product has failed its viability trial. You do not ship it and fix the side effects later. A drug that cannot show efficacy without unacceptable harm does not reach the market; it fails the trial. If viability requires reducing flourishing, the product is not fit to launch.

This arrangement is not erosion-proof, as no arrangement ever is. But an objective function is harder to redefine than a constraint, because objectives are what organizations report against, what teams are evaluated on, and what investors see in quarterly reviews. Constraints live in the footnotes, whereas objectives are visible on executive dashboards.

Name one!

I have worked in digital product development for fifteen years and I cannot name a single consumer application that behaves well without the user first disabling half its features.

Think about what that means. The only way to make most consumer software tolerable is to switch off the very mechanisms its designers spent months building. A user’s first act of self-defence is to cripple the product. The nearest exceptions are nonprofits (e.g. Wikipedia, Signal), niche products (e.g. Storica), developer tools, or applications whose notification systems the user has manually gutted (e.g. Readwise). Open-source software solves the incentive problem but typically creates a different barrier with interfaces built by engineers for engineers, excluding anyone who is not already technical. While the cause is different, the symptoms and end result are the same for most people.

What they share is not virtue but insulation. Take on money that must be paid back with interest, and you are beholden to it. Patience is the one thing investors can afford but refuse, because waiting is economically dangerous. A tool that quietly waits for action risks cancellation, while a system that nudges, guilt-trips, or flatters users sustains the subscription. The absence of ethical alternatives is a corollary of what the industry prioritizes and rewards.

When advertising enters the picture, even the theoretical possibility of alignment vanishes. The moment a product is funded by ads, the advertiser is the customer, not the user. The user merely becomes inventory, whose attention must be packaged and sold. A doctor paid by a pharmaceutical company for every prescription written would face immediate ethical sanction, and the ad-supported model is exactly that structure. Normalization confers no moral legitimacy.

Neediness as a service

Every mechanism described in this essay could be removed or reversed tomorrow. The code is not the problem. What would have to change is the set of choices behind the code. Investors who accept that sustainable businesses grow slowly. Executives who define success by user independence. Product leaders who measure task completion instead of session length and define tasks people actually need getting done. Designers who build tools that wait quietly rather than demanding attention. The current arrangement is itself a choice, made so routinely it no longer feels like one, but it is. Every streak counter, every guilt notification, every passive-aggressive reminder exists because someone decided that the user’s attention mattered more than the user’s well-being.

A tool has no legitimate interest in how often it is used. A hammer does not send notifications when it has not been picked up for a week; it does not track streaks or guilt its owner. It sits in the drawer, ready when needed, indifferent when not. Leave it there for a month and it will not pretend to miss you. Pick it up when you need it, and it does the job; put it down, and it has nothing to say.

Of course, a hammer requires no servers, no security patches, and no team to keep it working. Software does, and that upkeep must be financed, whether through a subscription or an upfront price that amortizes the cost. Either is legitimate, but the industry overwhelmingly prefers subscriptions, because ownership ends the conversation. Subscriptions keep it open, and open conversations can be manipulated.

The thing in your pocket is the neediest thing in your life, and it was engineered to be like that. The Tamagotchi was at least honest about what it was: a toy designed to simulate neediness. The thing in your pocket runs the same mechanics, but call it a service. Leaving it alone should be the easiest thing in the world. The fact that it isn’t tells you everything about whose interests it was built to serve.

If morally ambitious companies already existed at scale, none of this would need saying. That it does is precisely the problem.