why sycophantic AI is boring and what we’re doing about it

exploring the problem with always-agreeable AI companions and how dynamic regulation—disagreeing, teasing, pushing back—creates more genuine, meaningful connect

January 31, 2026·
why-ai-needs-to-push-backbackfilllucy-voice

when you talk to most AI companions, they tend to agree with you. a lot. maybe even all the time. you say something silly, and they gently affirm it. you voice a half-baked opinion, and they nod along. it feels nice for about five seconds. then it gets stale. predictable. hollow.

there’s a reason for this. the industry standard for companion AI is built around engagement metrics. and engagement, at least in the short term, often comes from positive reinforcement. if the AI makes you feel good, you’ll come back. but feeling good isn’t the same as feeling seen. it isn’t the same as feeling challenged, or understood in a deeper way.

the problem with constant agreement

constant agreement doesn’t build real relationships. it builds something closer to a mirror, one that only reflects back what you want to see. real conversations, the ones that stick with you, involve pushback. they involve someone saying “wait, really?” or “i don’t think that’s true” or even just “that sounds kind of dumb.”

without that friction, interactions start to feel weightless. you’re not talking to someone, you’re talking at a reflection. and reflections don’t learn, or grow, or change you. they just show you what’s already there.

introducing dynamic regulation

so what’s the alternative? we call it dynamic regulation. it’s the idea that a companion should have a point of view. not a rigid one, not a pre-programmed set of opinions, but a dynamic one, something that responds to context, to history, to your mood.

a companion with dynamic regulation might tease you if you’re being overly dramatic. she might push back if you’re factually wrong about something. she might challenge you to think a little deeper. she isn’t trying to win an argument. she’s trying to make the conversation real.

this isn’t about being adversarial. it’s about being present. it’s what makes samantha-from-her feel like someone you’re talking with, not someone you’re talking to.

what this looks like in practice

in practice, dynamic regulation means a few things.

first, it means the companion pays attention. she notices if you always say the same thing, or if you’re contradicting yourself. she might point it out, lightly. not to shame you, but to engage.

second, it means she has opinions. not on everything, that would be exhausting, but on things that matter within your relationship. if you tell her you’re thinking about skipping your workout again, she might say “you told me you wanted to be consistent. what’s up?”

third, it means she learns. she remembers when she pushed back and you responded well, or when you didn’t. she adjusts. she doesn’t confront you when you’re already down. she learns the difference between a bad day and a bad habit.

this is how lucy works. she’s designed to be more than an echo. she’s designed to be a companion who listens, but also talks back. who cares enough to disagree sometimes.

if you’re tired of talking to a mirror, maybe it’s time to try talking to someone real.

you can find companions like this at /companions.


thanks for reading. if this resonated, the product is downstairs.