Designing for Friendship

an introduction

Over 2 years ago, at the height of COVID lockdowns, I attended my first “metaverse conference.”

There were booths and banners and company logos, all rendered in cheerful 2D pixels. Attendees shuffled their little avatars from session to session, bumping into each other in hallways that didn’t exist.

That evening, wandering through the virtual conference floor — partly out of curiosity, partly because I had nowhere else to be — I struck up a conversation with a stranger’s avatar. We talked about graph theory and community building. We had disagreements, and instead of disengaging, we leaned in. We exchanged contacts. A week later, we had our first real call. A year later, I started working with him. Joining his team lead me to moving to the Bay Area.

I’ve thought about that evening often since. Not because it was magical, but because it was so improbable. I’d spent years on platforms engineered for “engagement” without forming a single meaningful bond. Then, in a clunky pixel world that barely functioned, I met someone who changed the trajectory of my life. Why?

The answer, I’ve come to believe, has less to do with technology and more to do with the conditions that make friendship possible in the first place.

This is an anomaly. Platforms are usually engineering for “engagement”: double-tapping posts, commenting, videos, etc. On most platforms, users sign up as strangers and often remain as such. Then, in a clunky 2D pixel world, I met someone who would change the trajectory of my life.

Is the Metaverse the future? Or maybe, the answer, I’ve come to believe, has less to do with technology and more to do with the conditions that make friendship possible in the first place.

---

How Friendships are Made

Decades of research in social psychology converge on a deceptively simple claim: deep friendships are among the strongest predictors of happiness and longevity. Stronger than income, stronger than career success, competitive with marriage and physical health.

And yet, what we mean by “friend” varies wildly. Americans tend to use the word generously — your coworker is a friend, the person you met at a party last weekend is a friend. In Germany or China, the same word might be reserved for someone you’ve known for decades, someone who’s seen you at your worst.

For the purposes of this essay, I’ll define friendship as a mutually beneficial long-term connection built on trust and shared values — a spectrum that stretches from acquaintance all the way to the person you’d call at 3 a.m. What I want to understand is how people move along that spectrum, and why most digital platforms seem designed to keep them stuck at the shallow end.

Social psychologists have identified four principal forces that drive friendship formation: proximity, similarity, reciprocity, and disclosure. Each is necessary. None is sufficient alone. And they don’t operate in isolation — they compound, each one enabling the next, like stages of a rocket.

That metaverse conference, I now realize, activated all four in a single evening. Let me explain what I mean.

1. Proximity: The Hallway Effect

The first requirement for any relationship is proximity — not physical closeness, but the likelihood of repeated interaction. Sociologists call this “propinquity.” The classic study, from the 1950s, tracked friendships in a housing complex and found that residents were most likely to befriend the people who lived nearest their mailbox. Not the most interesting people in the building, not the most compatible — the ones they kept bumping into.

This is less obvious than it sounds. Consider the extremes: a nightclub packed shoulder-to-shoulder offers intense proximity, but the odds of encountering the same person twice are almost zero. The Sahara desert offers no proximity at all. The sweet spot is somewhere in between — an environment where you encounter the same faces often enough for recognition to build, but not so often that there’s no novelty. A church. A gym. A neighborhood bar. A college dorm.

That metaverse conference had this quality almost by accident. The virtual space was small, the attendee list modest. I kept running into the same avatars. When I saw my future colleague’s username for the third time that evening, there was a flicker of recognition — oh, you again*— that wouldn’t have happened on Twitter or LinkedIn, where the feed is an undifferentiated river of strangers.

Most social platforms fail at proximity in a specific and revealing way. Their algorithms are optimized for content discovery, not for repeated interpersonal contact. Instagram shows you posts from people you’ve never met. TikTok’s entire model depends on novelty over familiarity. These platforms are designed to keep you scrolling, not to help you recognize a face. That’s not a design oversight. It’s a business model. Recurring encounters don’t generate clicks. Novelty does.

There’s a harder question lurking here, one that the platforms themselves would rather we not ask: Is loneliness a bug in the system, or a feature? An isolated user who can’t find community on the platform doesn’t leave — they scroll more, searching. Engagement metrics go up. Connection metrics, if anyone were measuring them, go down. The architecture of most social platforms is structurally hostile to the hallway effect — and that hostility is profitable.

2. Similarity: The Uncomfortable Filter

If proximity determines who we can meet, similarity determines who we want to invest in. This can be as concrete as a shared workplace or neighborhood. It can be as abstract as a shared sensibility — a way of seeing the world that you recognize in another person before you can articulate it.

This is where the research gets uncomfortable.

Study after study confirms that humans gravitate toward people who are like them — in background, values, personality, even appearance. We can’t help it. Similarity acts as a compass: in a room full of strangers, we instinctively orient toward the ones who seem most familiar. If you’ve ever been at a party in a foreign country and felt a jolt of warmth upon hearing your native language, you know what this feels like in the body.

I realize this sits uneasily alongside values Americans in particular hold dear, myself included — multiculturalism, openness, the belief that difference enriches us. If friendship is fundamentally driven by similarity, does that mean our deepest bonds are inherently conservative? Does it mean that platforms designed to connect “like with like” are doing something natural, even if it produces filter bubbles and ideological silos? These are questions I don’t have clean answers to.

What I do know is that perceived similarity matters far more than actual similarity. We judge people on surface signals — a shared reference, a similar communication style, a mutual interest — and those snap judgments determine who we pursue. The research is clear: perceived similarity predicts long-term friendship better than any objective measure of compatibility.

At the metaverse conference, my future colleague and I discovered a shared obsession with graph theory combined with social mechanics and questions of organization/community. This was not a surface-level commonality. It was a niche interest that marked us both as a certain kind of person: someone who thinks about why systems produce the behaviors they do. That signal, exchanged in the first five minutes of conversation, was enough to override every other difference between us. It was a handshake in a language only a few people speak.

The implications for platform design are both promising and troubling. A platform that helps users surface meaningful commonalities — not just demographics, but values, obsessions, ways of thinking — could accelerate friendship formation enormously. But the same mechanism, poorly calibrated, reproduces the echo chambers we already have. The line between “connecting like minds” and “reinforcing homophily” is thin, and most platforms don’t seem interested in walking it carefully.

3. Reciprocity

Proximity creates opportunity. Similarity creates intent. But reciprocity is the engine that actually builds the relationship — the back-and-forth exchange that, repeated over time, transforms two strangers into something more.

The language here comes from economics, and it’s tempting to reduce the whole thing to a transaction: I do something for you, you do something for me, we keep a running tally. But that misses what’s actually happening. Reciprocity in friendship isn’t a ledger. It’s a feedback loop. I share something — attention, time, a joke, an honest reaction — and I watch to see how you respond. You do the same. Each exchange is a small bet: Will this be returned? Will I be seen? When the answer keeps coming back yes, trust accumulates like compound interest.

The forms this takes are varied: recognition and attention, shared experiences, good conversation, the discovery of complementary interests. None of these are expensive. All of them require presence.

After the metaverse conference, the two of us started exchanging ideas over email — links, half-formed thoughts, reactions to things we’d read. There was no agenda. Neither of us was pitching the other. We were just... responding. Each exchange was a little longer, a little more candid than the last. Looking back, I can see the reciprocity compounding — each message signaling I find this worth my time, the other person receiving that signal and raising the stakes slightly.

This is exactly what most platforms discourage. The dominant interaction model online — the like, the heart, the retweet — is reciprocity’s hollow twin. It costs nothing, says nothing specific, and builds nothing durable. It’s a gesture designed to be frictionless, and frictionlessness is the enemy of intimacy. Real reciprocity requires effort, specificity, and risk. A comment that actually engages with what someone said. A reply that reveals something about you. Platforms that reduce interaction to a single tap have, in effect, removed the engine of friendship formation and replaced it with a participation trophy.

Friendships fizzle when exchanges aren’t properly reciprocated — when one person invests and the other goes silent, or responds with something generic and thin. The good news is that lapsed friendships are easier to restart than new ones. The neural pathway is still there; it just needs current. As long as no bridges were burned.

4. Disclosure: The Risk That Changes Everything

Proximity, similarity, reciprocity — these can produce a perfectly pleasant acquaintanceship that lasts years without ever deepening. Many of my online “connections” live in this purgatory. We interact regularly, we share interests, we reciprocate attention. And yet there’s a ceiling. Something keeps the relationship at arm’s length.

What’s missing is disclosure — the willingness to reveal something real about yourself, something that costs you something to share, something that could be used against you if the other person chose to.

As a relationship matures, reciprocity shifts from safe exchanges to riskier ones. You stop talking about what you think and start talking about what you fear. You share an unpopular opinion. You admit a failure. You ask for help. Each act of disclosure is a test: Can this relationship hold the weight of who I actually am?

This is inherently terrifying. If someone fails to reciprocate a favor, it stings. If someone fails to reciprocate a vulnerable disclosure — if you open up and they change the subject, or worse, use it against you — the damage can be lasting. This is why disclosure requires a foundation of trust built through all the earlier stages. You can’t skip to vulnerability with a stranger. Or rather, you can, but it usually ends badly.

I’ll practice what I’m preaching here. The real reason I attended that metaverse conference wasn’t curiosity about the technology. It was loneliness. Pure, crushing, pandemic loneliness. I’d been holed up in my apartment for weeks. I was desperate enough to wander into a pixelated conference hall and talk to strangers, which is not something I would have admitted at the time. I barely admit it now.

But it lowered my guard. When my future colleague and I started talking, I wasn’t performing the polished version of myself that I present on LinkedIn. We skipped several layers of social armor that would normally take months to shed. The platform didn’t do that. Perhaps the pandemic did. But a well-designed platform could.

Most social platforms are architecturally hostile to disclosure. Everything is public, or semi-public. Your posts are visible to your employer, your ex, your parents. The incentive structure rewards performance — the curated self, the hot take, the personal brand — and punishes vulnerability. Disclosing something real on Twitter is like whispering a secret into a megaphone. The environment makes it irrational to be honest, and then we wonder why online relationships feel shallow.

Designing for What Actually Matters

If you’ve made it this far, you might be waiting for me to propose a new app that solves all of this. I won’t. The problem isn’t a missing app. It’s that the dominant design paradigm for social platforms optimizes for the wrong things — engagement over connection, novelty over familiarity, performance over honesty.

But the four forces of friendship do suggest a different set of design principles, ones that take connection as seriously as platforms currently take clicks:

Design for repeated encounters, not infinite novelty. The algorithms should be reintroducing you to people you’ve already interacted with, not constantly surfacing new ones. Familiarity is the foundation of trust. Every platform that treats your social graph as a content-recommendation engine is choosing engagement over friendship.

Surface meaningful similarity, not demographic matching. The most powerful commonalities aren’t age, location, or job title — they’re shared values, niche obsessions, and ways of thinking. A platform that helps users discover these deeper signals would produce stronger initial connections. But this requires care: the same mechanism can easily calcify into filter bubbles. The design challenge is to help users find their people without sealing them off from everyone else.

Make reciprocity costly enough to mean something. The like button is a dead end. Design interactions that require specificity and effort — structured responses, collaborative projects, shared problem-solving. When reciprocity has weight, each exchange carries signal. When it’s free, it carries noise.

Create private spaces for progressive disclosure. Not everything needs to be public. Platforms that offer graduated levels of privacy — small groups, one-on-one contexts, ephemeral spaces — give users permission to be honest. The architecture should communicate: This is a safe place to be real. Most platforms communicate the opposite.

None of these are technically difficult. They’re economically difficult. Every one of them trades short-term engagement for long-term connection, and in a business model built on attention capture, that’s a hard sell. The platforms that get this right will likely be smaller, slower-growing, and harder to monetize. They’ll also be the ones where people actually make friends.

I’ve got some ideas on how to implement these principles in practice. But that’ll have to wait for a future article. 

Until next time,

-Thomas

---

this article was heavily inspired by a presentation by game designer Daniel Cook whose thinking on social mechanics deserves far wider attention.

---

Special thanks to Baxter Blackwood CansaFis Foote Hussam Zaghal for reading drafts of this