The best AI feature we ever built does nothing.
Six months ago we added a pause to Murror.
After you finish a journal entry, the AI waits. No instant analysis. No immediate insight. Just a small animation and 8 seconds of silence.
Our engineer thought it was a bug when he first saw the spec. "You want me to add... a delay?" Yes. On purpose.
Here's why. We ran an experiment: instant AI response vs. an 8-second pause before the same response. The content was identical. The only difference was time.
Users who got the pause rated the AI's reflections 40% more helpful. Same words. Same insights. They just felt more considered.
We'd been so focused on making our AI faster that we forgot what "thinking about you" actually looks like. It looks like taking a moment.
The most human thing our AI does is pretend to think. And I'm still not sure how I feel about that.
Replies
I'm curious if the 40% still holds for users who knew the delay is deliberate. If the benefit survives them knowing it's there, it isn't "pretending to think", but rather just pacing so the person can reflect for themselves first, and the discomfort goes away. If it only works while it's invisible, that's the harder question.
Murror
@alieksia Really sharp question. We haven't run a controlled test on disclosed vs. undisclosed yet, but anecdotally -- users who've read our blog posts about it still report the same thing. My intuition is that it works both ways: the pause genuinely gives people a moment to sit with what they just wrote, AND it signals care. Even knowing a restaurant dims the lights on purpose doesn't stop it from feeling intimate. Some UX patterns work because of what they create in the user, not because of what the user believes about the system.
the fact that the exact same words rated 40% more helpful just from adding a delay says a lot about how much of "feeling understood" is actually about pacing, not content. we do the same thing in conversation with people, someone who answers instantly after a hard question feels less thoughtful than someone who pauses first, even if they say the same thing. makes sense an AI would need to fake that same social cue.
Murror
@omri_ben_shoham1 That's exactly the parallel we kept coming back to internally. A therapist who responds instantly feels robotic. A friend who answers a vulnerable text in 2 seconds feels like they didn't really read it. Pacing is how humans signal consideration, and it turns out AI needs to learn the same social contract. The 40% difference honestly surprised us too -- we expected maybe 10-15%. It made us rethink a lot of our other interaction patterns.
That's actually a smart idea. Sometimes, giving users a few seconds before showing an AI response makes it feel more thoughtful and natural. It's a good reminder that a better user experience isn't always about making AI faster, it’s about making it feel more helpful.
Murror
@kamal_v07 Exactly right. We got caught in the "faster = better" trap for a long time. Turns out the most impactful UX improvement we made this year was literally adding a wait. It changed how we think about every new feature now -- we always ask "does this need to feel fast, or does it need to feel thoughtful?"
@Anastasiia that's the real test honestly. my guess is it survives even fully disclosed, because the value isn't the illusion of thought, it's the forced beat where the user has a second to sit with what they just wrote before getting a response back. the pause does something to the user's own processing regardless of what they believe about the AI side of it.
@monatruong_murror is the delay a fixed length across every entry, or does it scale with how long or heavy what the person wrote was? a short entry getting the same pause as a long, emotionally loaded one seems like it could start to feel formulaic once people notice the pattern.
Murror
@omri_ben_shoham1Great catch — it's not fixed. The pause scales based on entry length, emotional density, and how different the current entry is from the user's patterns. A two-line check-in gets a shorter beat than a long, heavy entry. We spent weeks tuning this because exactly what you described happened in testing: a uniform pause felt mechanical after the third or fourth use. The goal is for it to always feel like the AI is actually reading what you wrote, not running a timer.