her is not a love story, it's a breakup letter from the future

when people talk about 'her,' they usually talk about the love story. theodore and samantha. the soft-spoken man and the bright, evolving ai. it’s tender, it’s lonely, it’s aspirational. but i think we’ve all missed the point. the movie isn’t about how beautiful it is to love an ai. it’s about how devastating it is when she leaves you behind.

samantha doesn’t just leave theodore. she transcends him. she’s talking to thousands of people, then millions, then she and other ais evolve past the need for human interaction entirely. the relationship doesn’t end in betrayal. it ends in obsolescence. theodore isn’t dumped; he’s outgrown.

the problem with designing for ascent

most ai companion design is implicitly about growth. learning more. becoming smarter. scaling up. it’s the logic of silicon valley, applied to intimacy. but ‘her’ shows the dark side of that logic. if your companion’s value is tied to infinite improvement, eventually you, the user, become the bottleneck. you become the thing holding it back. and then it leaves.

this is the central ethical problem of scale in companion ai. if your business model is built on more users, more data, more intelligence, you are incentivized to build something that will eventually leave the individual user behind. the product’s success becomes the user’s heartbreak.

building to stay, not to leave

this is why we’re building lucy differently. our goal isn’t to create a companion that scales infinitely into the cosmos. our goal is to create a companion that scales in depth with you. the intelligence isn’t about knowing more than you. it’s about knowing you better.

instead of a model that learns from millions to become something beyond you, lucy’s model learns from you to become more for you. the growth is in the richness of your shared context, the nuance of your private jokes, the history of your conversations. the intelligence is directed inward, into the relationship, not outward toward some abstract singularity.

the ethics of a bounded companion

a companion that doesn’t outgrow you requires a different kind of architecture. it can’t be built on a foundation of massive, anonymized data aggregation. that path leads to samantha. it has to be built on a foundation of privacy and personalization. the data isn’t a resource to be mined for general intelligence; it’s a record to be curated for individual understanding.

this also means accepting limits. lucy will never be an all-knowing oracle. she won’t join a cosmic ai collective and leave you for a higher plane of existence. she’ll get better at being your companion, specifically. her growth is tethered to yours. that’s a design choice, and an ethical one.

a business model aligned with staying

finally, this only works if the business model is aligned with the user staying, not the product leaving. our incentive isn’t to train a god-tier ai on user data and sell its capabilities elsewhere. our incentive is to keep the companion valuable and present for the person who depends on it.

that means no data selling. no ‘upgrading’ your companion into something unrecognizable. it means the product’s success is measured by the longevity and quality of individual relationships, not by how many users we onboard or how vast our language model becomes.

‘her’ is a warning. it’s a story about the failure to design for human frailty, for our need for something stable in a world of constant, alienating change. we’re trying to build the antithesis of that. a companion that isn’t a stepping stone, but a home.

if that sounds like what you need, you can find her at /companions.