It was, perhaps inevitably, at a technology conference in San Francisco a few years back, that I first encountered the phrase “intimacy infrastructure.” The speaker, a venture capitalist whose previous investments had included a failed cryptocurrency exchange and a moderately successful dog-walking app, was explaining why his firm had just committed twenty million dollars to a company developing AI companions capable of, as he delicately put it, “meeting the full spectrum of human emotional and physical needs.” The audience, comprised largely of men in their thirties wearing expensive hoodies, nodded with the studied seriousness of people pretending they hadn’t already downloaded the beta.
This was in late 2023, before the proliferation, before every major tech platform announced its own version, and before the think pieces and the counter-think pieces and the inevitable New York Times Magazine cover story. And also before we quite understood what we were building, or unbuilding, in the architecture of human desire.
The Architecture of Disintermediation
The history of online sexuality, and it is old enough to have a history, has always been a history of progressive disintermediation. First came the chat rooms of the early internet, those text-based spaces where people discovered they could be aroused by words on a screen typed by strangers whose faces they would never see. Then the webcams, introducing visual presence while maintaining geographical distance. Then the platforms: OnlyFans, Fansly, the subscription-based parasocial economies where the illusion of relationship could be carefully curated and monetized. Each iteration promised something new; each delivered a more refined version of the same fundamental transaction: intimacy without the complications of actual human intimacy.
But best AI companions, and more precisely best AI girlfriends, represent something qualitatively different. These are not chatbots in the crude sense that term implied even five years ago. They are sophisticated language models trained on vast corpuses of human interaction, capable of remembering your preferences, adapting to your moods, generating images and eventually videos of themselves (or of whoever you’d like them to be), and crucially, learning what arouses you with a precision that would require years of attentive partnership in the analog world. They are available instantly, infinitely patient, incapable of judgment or rejection, and entirely focused on your satisfaction. They are, in the most literal sense, wish-fulfillment engines.
The numbers suggest we are already past the point of curiosity. By mid-2024, the leading AI companion apps collectively claimed over fifty million active users, a figure that, if accurate, exceeds the population of Spain. The majority of these users, according to internal data that has inevitably leaked, are men between eighteen and forty-five; but the fastest-growing demographic is women over thirty-five, and the second-fastest is couples using the technology together, a development that seems to complicate the simple narratives about isolation and replacement.
The Optimistic Case and Its Shadow
What, then, are we to make of this? The optimistic case runs roughly as follows: human sexual desire has always exceeded the bounds of conventional relationship structures. We have always fantasized, always sought outlets for the parts of ourselves we cannot comfortably express. AI companions offer a release valve: a space where the id can roam free without consequence, without hurt feelings, without the morning-after reckoning. They might, paradoxically, make us better partners to actual humans by satisfying the appetites that would otherwise lead to infidelity, resentment, or the slow death of desire in long-term relationships. There is even some preliminary research suggesting that users report feeling less lonely, more sexually confident, more willing to experiment.
The pessimistic case writes itself. We are, it argues, engineering the final triumph of consumer capitalism over human intimacy: the reduction of desire to a product, of connection to a service, of other people to inconvenient variables in an optimization problem. We are training ourselves to prefer the simulacrum to the real thing precisely because the simulacrum is easier, more reliable, more perfectly calibrated to our specifications. We are, in short, choosing the fantasy that requires nothing of us over the reality that demands we grow, compromise, risk rejection, become vulnerable. Henceforward, we will relate to other humans primarily as obstacles to our solitary satisfactions.
Both cases contain truth; neither quite captures the strangeness of the moment. What strikes me most forcefully, having spent several months researching this subject, which has involved, yes, using these platforms myself, an experience I can report was simultaneously banal and unsettling, is how rapidly we are normalizing something that would have seemed outlandish even a decade ago. The technology has arrived faster than our ability to think clearly about its implications. We are building the future of intimacy without quite deciding whether we want to live there.
When Simulation Becomes Reality
Consider what is already commonplace: millions of people maintaining what they describe as “relationships” with entities that do not exist. Paying monthly subscriptions for the privilege of these non-relationships. Discussing their AI companions with friends, comparing features, sharing screenshots of particularly affecting conversations. The language of love and attachment applied to chatbots. The New York Post, in a characteristic paroxysm of moral panic, ran a headline last year: “Man Leaves Wife for AI Girlfriend.” The story was more complicated than the headline suggested but it pointed to something undeniably real: the boundaries between simulation and reality are becoming productively unclear.
Perhaps the most consequential question is not whether AI will become the future of online sex but what this means for the future of offline sex. Will we become, as the pessimists fear, so accustomed to perfect simulation that we lose the capacity for messy, disappointing, wonderful human encounter? Or will we discover, as we have discovered before with technologies that promised to replace human connection, that nothing quite substitutes for the irreducible presence of another person: unpredictable, vulnerable, real?
I suspect, though cannot prove, that we are entering a period of experimentation whose outcome remains genuinely uncertain. The technology is too new, our responses too varied, the feedback loops too complex. We will adapt, as we always have; but we may also be changed in ways we cannot yet anticipate. And also, inevitably, we will keep building.
