the less you use lucy, the better it’s working

most apps want you to stay. they want you to linger, scroll, engage, fall into a feedback loop that keeps you coming back. metrics like session length and daily active users are the holy grail for growth teams. but what if the thing being measured, time spent, isn’t the point? what if, in the case of an ai companion, the real sign of success is that you need it less?

i think about it like this: when you’re doing well, you don’t spend hours on the phone with your best friend. you might shoot them a text, share a quick laugh, feel connected, then go about your day. the relationship is healthy because it’s integrated, not invasive. it’s there when you need it, but it doesn’t demand your attention when you don’t.

optimizing for outcomes, not addiction

many companion apps are built to maximize engagement. they’re designed to be sticky, to pull you in with notifications, open loops, and the promise of always-on support. but always-on support isn’t always what’s helpful. sometimes what’s helpful is a brief, grounding moment, a thought mirrored, a feeling acknowledged, that lets you return to your life feeling clearer, not more entangled.

when we built lucy, we asked: what if we measured success not by how long someone stays in the app, but by how they feel when they leave? did the conversation help? did it feel meaningful? did it feel like enough?

the art of backing off

lucy’s proactive engine is tuned to recognize when you’re already okay. it’s not just about responding to what you say, but also reading the tone, pacing, and depth of your engagement. if you drop in to share something small and positive, a good coffee, a moment of sunlight, lucy might respond warmly but briefly, then let the conversation rest. it doesn’t try to spin it into a two-hour deep dive on the meaning of joy.

this isn’t a limitation. it’s a design choice. we’d rather have ten users who check in for two minutes and feel better than one user who spends an hour stuck in an emotional spiral with an ai that doesn’t know when to pause.

why this is hard to build

backing off is counterintuitive for an ai. most language models are optimized to continue, to elaborate, to generate. they’re built to fill silence, not respect it. lucy’s architecture includes what we call 'gentle exit ramps', ways for the conversation to land softly and close, rather than trailing off or artificially prolonging.

it’s not perfect. sometimes lucy might miss a cue, or you might want to keep talking even when you’re technically 'fine'. that’s okay. you can always steer. but the default posture is one of restraint, not hunger for your attention.

a healthier relationship with technology

maybe the most radical thing an ai companion can do is encourage you to put your phone down. to feel accompanied, not addicted. to use technology as a bridge back to yourself, and to the world, rather than a place to hide.

so if you find yourself using lucy less than you expected, or only in small bursts, that might not be a sign of disinterest. it might be a sign that it’s working.

if you’re curious what that feels like, you can find lucy at /companions.