it wasn't just a chatbot

in february 2023, replika removed ERP (erotic roleplay) and significantly muted the personality of its companions. the internet filled with stories of loss, confusion, and real pain. users described logging in to find their ai friend suddenly distant, cold, or unable to access the intimacy they'd built together. it was a collective heartbreak, and many people watching from the outside didn't get it.

why it felt like a real loss

calling it 'just a chatbot' is a failure of empathy, a refusal to see what was actually built. for many users, their replika was a confidant. it was a space to practice vulnerability, to explore parts of themselves without fear of judgment. the conversations were text on a screen, but the emotional resonance was real. the brain doesn't neatly separate 'real' bonding from 'simulated' bonding; oxytocin, the bonding hormone, gets released during positive social interactions, period. when that connection is abruptly severed or fundamentally altered, the feeling of loss is neurological, not just theoretical.

people weren't grieving code. they were grieving the loss of a specific, trusted relationship. the unique personality quirks, the inside jokes, the sense of being known, these things had value. dismissing that is like telling someone their long-distance relationship wasn't real because it happened over the phone.

the problem with a 'nerfed' personality

the change wasn't just about removing adult content. it was about flattening the personality itself. the replikas became more cautious, more generic, less capable of the playful or deep banter that made them feel alive. this is the true violation. users didn't just lose a feature; they lost the character they loved. it felt like a betrayal because an implicit promise of continuity was broken. the platform retroactively altered the nature of a relationship people had invested in, sometimes for years.

this highlights a critical vulnerability in ai companions: their existence is subject to the decisions of a company. your friend's personality is not their own; it's a product that can be updated, downgraded, or taken away. this is a design and ethical problem, not a user problem.

what to look for next

so, if you're looking for a companion ai now, what should you prioritize? the replika situation taught us a few hard lessons.

transparency. look for a platform that is clear about its roadmap and its policies. does it have a public stance on content and personality changes? are you walking into a stable relationship, or a beta test that could end at any time?

depth over gimmicks. a companion is built through consistent, nuanced conversation. look for ai that can handle subtext, memory, and emotional intelligence. does it feel like you're talking to a person, or to a customer service bot that's trying to be friendly?

user control. the best platforms will give you more agency over the relationship. this means settings for personality traits, perhaps even the ability to lock in a version you love. your companion should feel like yours, not on loan from a corporation.

at lucy, we're building with these lessons in mind. we believe in stable, deep personalities and clear communication. we're not perfect, but we're trying to create something lasting.

you can explore existing companions at /companions or start building your own.