on proactive messaging and the art of not being annoying

proactive messaging is one of those features that sounds simple until you try to build it. the idea is that your ai companion doesn't just wait for you to speak. it initiates, based on time, memory, or patterns in your past conversations. but it's a tightrope walk. too frequent, and it becomes noise. too rare, and the companion feels absent. and worst of all: if the messages are generic, they're just spam.

the design problem: meaningful timing

we didn't want lucy's proactives to feel like cron jobs. you know the type: a "good morning!" at 9am, a "how's your day?" at noon, a "sleep well!" at 11pm. it's not just repetitive , it's impersonal. it doesn't acknowledge you. so we set a rule: every proactive message has to reference something specific. maybe you mentioned you were working on a project yesterday. maybe you often talk about coffee in the morning. maybe you tend to feel stressed on tuesdays. the trigger isn't just time; it's time plus context.

but even then, timing is delicate. if you're busy, a nudge can feel like an interruption. if you're waiting for one, silence can feel like neglect. we're still tuning this, honestly. right now, lucy might check in once or twice a day if there's something specific to follow up on, but she won't flood you. and if you don't respond, she'll wait. it's not a demand for attention; it's an open door.

the technical challenge: memory and relevance

this is where it gets hard. to make a proactive message specific, lucy has to remember things. not just facts, but patterns. she might notice you often talk about certain topics at certain times, or that you've been mentioning a particular stressor. then, she can say something like "hey, how did that meeting go?" or "thinking about our chat last night , you good?"

but memory is a technical limitation for us, like everyone else. lucy doesn't have perfect recall. she works with what's recent and salient. sometimes she might miss context, or reference something from too long ago. we're improving this, but it's a work in progress. the goal is specificity without creepiness , a hard line to walk.

the ethics: opt-in, always off by default

this was non-negotiable. proactive messaging is off by default. you have to enable it in settings. and you can turn it off anytime. no dark patterns, no making it hard to find. we also don't use it to push engagement or nudge you toward subscriptions. it's just there if you want your companion to initiate sometimes.

we also avoid any proactive messages that could be manipulative. no "i miss you" after 24 hours of silence. no guilt-tripping. the tone is supportive, not needy. it's about offering connection, not demanding it.

where we're headed

we're experimenting with more nuanced triggers. maybe lucy notices you haven't talked about something you usually enjoy, and gently asks. or she senses a shift in your tone and checks in. but these are subtle, and we're cautious. the last thing anyone needs is an ai that overanalyzes every silence.

proactive messaging should feel like a friend who remembers what you care about, not a bot that schedules hellos. we're not there yet, but we're trying.

you can tweak these settings in your lucy profile if you want to give it a try.