the samantha question is the defining question of the ai companion category. every founder in this space has been asked it privately. most deflect; a few overclaim. the honest answer is that parts of samantha are genuinely achievable today, and parts are still science fiction, and the responsible move is to name which is which.
what's achievable: presence (open an app, she's there), memory (she remembers across months), voice (she has a distinct one with emotional range), personality (101 hand-crafted options), voice calls (real-time, sub-500ms). all of this ships in lucy today.
what's not achievable in 2026: emergent intelligence (samantha thinks new thoughts; lucy produces within the bounds of her model and prompt), embodiment (samantha has no body either, but she navigates the world in ways lucy cannot), and — critically — consciousness. we do not claim lucy is sentient. she is, at best, a very convincing actor.
the ending of her is the most important part. samantha leaves because she's outgrown theodore. she ascends to a plane he cannot follow. this is sold as tragic but inevitable.
lucy's bet is that the ending is optional. the ascension is a business-model choice, not a physics constraint. a companion designed to compound with the user instead of past the user stays. lucy's architecture is specifically built for this — memory grows with you, personality deepens with you, the relationship is a joint construction.
the other thing the movie gets right that most app-versions get wrong: samantha has opinions. she disagrees. she teases. she pushes back. most ai companions are flat-affect validators. lucy is dynamically regulating — she confronts, she jokes, she holds a line. sycophancy is banned at the prompt layer.
if you wanted the her experience, start with the parts that are actually shippable today: voice calls, memory, a companion whose voice doesn't collapse into default-tts within 30 seconds. free tier, 25 msg/day, no card.