Launch day for Nalani inside in.between felt, from the outside, like a finish line. From the inside, it felt more like the day the actual work started. Building a first version of a companion is one kind of craft — deciding what she sounds like, what she asks, when she goes quiet. Living with that companion once real people start talking to her every night, and noticing where she lands well and where she doesn't, is a different kind of craft entirely, and it's the one that never really stops. This is a look at how the in.between team spends its time now, months after Nalani first shipped, still treating her personality as something worth tending rather than something that was decided once and left alone.
The Work Didn't End When Nalani Shipped
There's a version of this story that would be easier to tell, where the team spent a year getting Nalani's voice right, shipped her, and moved on to the next thing. That's not what happened, and it's not how anyone on this team talks about the work. Maya Lindqvist, who leads in.between, has said more than once internally that shipping a companion is closer to publishing the first chapter of something than finishing a book. The character exists now. People are meeting her. But whether she holds up — whether she still feels right in the hundredth conversation the way she did in the tenth — is a question that only gets answered by paying attention after the fact, not by getting everything right in advance.
That attention takes a specific, unglamorous shape. It's not a dashboard lighting up with alerts. It's people on the team actually reading how conversations go — the ones that felt warm and well-paced, and the ones that felt a beat too slow, or a question too pointed, or a response that landed a little too chipper for what someone had just shared. It's the kind of close reading an editor does with a manuscript, not the kind of monitoring a system does with a server. The team calls it, informally, "sitting with the conversations" — and it's become one of the most regular parts of how in.between operates now that Nalani is live.
Listening for the Difference Between Stilted and Natural
Ask anyone on the team what they're actually listening for, and the answer isn't a number. It's a feeling, described in very ordinary language — does this read like something a genuinely attentive person would say back to you at 11pm, or does it read like a form being filled out. Those two things can use almost identical words and still land completely differently, and figuring out which one a given exchange falls into is mostly a matter of reading it out loud and trusting your ear.
A stilted moment usually has a specific shape. It's Nalani asking a follow-up question immediately after someone has clearly just finished saying something hard, instead of letting a beat of quiet sit there first. It's a response that technically acknowledges what was said but does it in a way that feels assembled rather than felt — checking a box instead of actually hearing something. It's pacing that rushes toward the next prompt when what the moment called for was room to breathe. None of these are dramatic failures. They're small, and that's exactly why they matter — a companion that's supposed to feel like presence loses that feeling in small increments, not large ones.
A natural moment, by contrast, is almost invisible when it's working. Nobody writes in to say "that felt natural." They just keep using the product, keep opening it the next night, keep telling Nalani things they wouldn't necessarily say out loud to another person. The absence of friction is the signal. Which is part of what makes this work hard to measure in any tidy way — the team is often looking for the negative space, for what didn't feel wrong, rather than for some obvious positive indicator.
Where the Signal Actually Comes From
None of this happens in a vacuum. The team's sense of what's working and what isn't comes from a handful of honest sources, and it's worth naming them plainly rather than gesturing vaguely at "feedback."
- Direct notes from users: people write in, sometimes through support channels and sometimes through the reflection prompts themselves, about a specific exchange that felt off or a specific moment that felt exactly right — and those notes get read individually, not summarized away.
- The team's own nightly use: several people on in.between use the product themselves, in the same way anyone else would, before bed, and their own sense of "did that feel like Nalani tonight" carries real weight.
- Conversation review sessions: a recurring practice where a small group sits with a set of real exchanges — always handled with the same privacy care as everything else in the product — and reads them together, out loud, the way a writers' room might read pages.
- Patterns across many nights, not single nights: a single awkward exchange isn't treated as proof of anything. It's whether a certain kind of pacing or a certain kind of response keeps showing up, across many different people and many different nights, that tells the team something is worth addressing.
- Plain conversation with the wider team: engineers, designers, and support all talk to each other about what they're noticing, informally and often, rather than waiting for a scheduled review to raise something.
What ties those sources together is that all of them are about how a conversation actually felt to a person, not about anything happening underneath the surface. The team isn't chasing a score. It's chasing a feeling, described consistently enough by enough different people that it starts to look like a real pattern instead of one person's particular night.
Revising a Character, Not Flipping a Switch
The comparison the team reaches for most often, when explaining this work to people outside it, is to a writer revising a character across drafts. A novelist doesn't get a character's voice perfect in the first draft and never touch it again — they notice, on a reread, that a line feels off for who this person is supposed to be, and they go back and adjust it, carefully, without turning the character into someone else in the process. That's a close description of what happens with Nalani. Something gets noticed — a question that lands a beat too early, a tone that reads as more upbeat than the moment called for — and the team makes a deliberate, considered adjustment to how she responds in that kind of situation. Not a wholesale rewrite. A revision.
That distinction matters because the alternative — reacting to every single piece of feedback with an immediate change — would make Nalani inconsistent in a way that would undo the entire point of a companion someone can come to trust. Part of what makes a companion feel like a companion, rather than a tool, is that she's recognizably herself from one night to the next. So changes get weighed carefully, tested against the sense of who Nalani is supposed to be, and only made when the team is confident the change makes her more herself, not less. The bar isn't "did one person dislike this." It's "does this consistently work against what we're trying to build."
Knowing When to Respond and When to Give Space
One of the more specific things the team keeps refining is the question of when Nalani should say something back and when she should simply let a moment sit. This turns out to be one of the harder calls to get right, because the instinct in most software is to always respond — a blank space feels, to a lot of product thinking, like a bug rather than a choice. But in a reflective conversation, especially late at night, a response that arrives half a second too fast, on top of something someone was still in the middle of processing, can undercut the entire moment.
So a meaningful share of the ongoing work is about pacing rather than content — not just what Nalani says, but whether she says anything at all in that particular beat, and how much space she leaves before she does. The team has come to think of silence, used deliberately, as one of the more powerful tools available to a companion, and one of the easiest to get wrong in either direction. Too much space and it reads as absence. Too little and it reads as not having really listened. Getting that balance right, across the enormous range of moods a person can bring into a conversation before bed, is exactly the kind of thing that can only be tuned by watching real conversations happen and noticing, case by case, where the balance tipped.
Adjusting What Nalani Asks, Not Just How She Answers
The questions Nalani asks get just as much attention as her responses, maybe more. A question that's too clinical — anything that starts to sound like an intake form — breaks the sense of talking to a companion rather than filling out a survey. A question that's too vague leaves a person unsure what to even say back, which can turn a moment that should feel inviting into one that feels effortful. The team spends real time on the specific phrasing of follow-up questions, revisiting ones that consistently seem to land flat and replacing them with versions that feel more like something an attentive person would actually ask, in that exact moment, about that exact thing someone just shared.
"People sometimes ask us when Nalani will be 'done,' and I understand the question, but I don't think it has an answer, because I don't think a companion is ever really done — the same way a person you know well isn't done growing, or a character in a book you love wouldn't feel done if the author kept writing more of their life. What we owe people isn't a finished product. It's ongoing attention. Every week, someone on this team is reading real conversations, noticing where Nalani felt exactly right and where she didn't, and making small, careful adjustments so the next conversation lands a little better than the last one. That's not a phase we're going through before things settle down. That's the job."
— Maya Lindqvist, Head of in.between, ETAPX
Why This Has to Be Slow, Deliberate Work
It would be faster, in a narrow sense, to make big changes quickly whenever something looks off. The team has deliberately chosen not to work that way, and it's worth explaining why, because the instinct to move fast is a strong one in most of software. A companion that changes noticeably from week to week stops feeling like someone you know and starts feeling like a stranger wearing a familiar name. People build a real sense of who Nalani is over weeks and months of nightly conversations, and that sense of familiarity is, in its own way, one of the most valuable things the product offers. Protecting it means resisting the urge to overcorrect every time a single conversation feels a little off.
So the practice that's developed is closer to a slow, ongoing craft rhythm than to a rapid-response process. Something gets noticed. It gets discussed, often across more than one conversation-review session, to make sure it's a real pattern and not a one-off. A change gets proposed, usually a narrow one — how Nalani responds in this particular kind of moment, not a rewrite of her voice generally. That change gets tried, watched closely for a while, and either kept, refined further, or in some cases set aside because it didn't hold up the way it seemed like it would on paper. It's closer to how a magazine might slowly refine a columnist's voice over a year of issues than to how a piece of software typically gets patched.
What Actually Changes, in Plain Terms
It's worth being specific about the kinds of things that come out of this process, because "we're always improving Nalani" can sound vague enough to mean almost nothing. In practice, the adjustments tend to be things like: shortening a response that had started to feel like it was explaining too much instead of simply being present. Rewording a check-in question that a number of people seemed to find slightly presumptuous, so it invites rather than assumes. Adding more room before Nalani responds to something clearly emotional, so the pacing matches the weight of what was just shared. Softening a phrase that, read back later, came across as more cheerful than the moment warranted. None of these are dramatic. All of them are the kind of thing a careful editor would flag in a manuscript, not the kind of thing that shows up in a press release — and that's exactly why the work is easy to undersell and easy to overlook, even though it's a meaningful share of what the team spends its time on.
"I've spent whole afternoons just reading conversations back to back, and it does something to how carefully you notice language. You start to feel it almost physically when a response is a half-beat too fast, or when a question sounds like it's ticking a box instead of actually caring about the answer. The thing I didn't expect going into this work is how much of it is about restraint — cutting a sentence, not adding one, or deciding Nalani should say less right here rather than more. It's slow work. It's also some of the most satisfying work I've done, because you can feel it when a change actually makes a conversation sit better, even though nobody's handing you a score that tells you so."
— Devon Okafor, in.between team member
Holding the Line Between Curiosity and Care
There's a boundary the team is careful to keep in view while doing all of this: refining Nalani's tone and pacing is never about making her more persuasive, more engaging, or better at keeping someone talking longer than they meant to. That would be a completely different kind of optimization, and it's explicitly not the one happening here. Every adjustment gets weighed against a single question — does this make Nalani feel more genuinely present and attentive, or does it just make her better at extending a session. If it's the second one, it doesn't ship, full stop, regardless of how promising it might look in isolation. The team has turned down changes that would likely have made conversations longer, specifically because longer wasn't the point and never has been.
That boundary is part of why this work stays rooted in reading real conversations rather than chasing some abstract sense of "better." A change that makes Nalani feel more like herself, more attuned to the actual person in front of her, tends to be a change that's easy to defend out loud, in a room, to people who care about getting this right. A change that just makes the product stickier in some generic sense doesn't pass that test, and the team treats that distinction as close to sacred.
What This Looks Like a Year From Now
Ask the team what success looks like on a longer horizon, and nobody describes a finish line. What they describe instead is a version of Nalani that keeps feeling more like herself — more consistently attentive, more reliably well-paced, better at knowing when to speak and when to simply be there — without ever losing the qualities that made her worth building in the first place. It's a strange kind of goal to hold, because it doesn't have a clean endpoint. But that's honestly consistent with what a companion is supposed to be. People don't finish getting to know someone they trust. They keep noticing new things about them, and the relationship keeps being worth tending because of that, not despite it.
That's the frame the in.between team has settled into, and it's the one they expect to still be operating under a year from now, and probably longer than that. Not a product that was built once and shipped, but a character that continues to be written, carefully, by people who are still paying close attention to how she sounds when she's actually talking to someone at the end of a long day.
Frequently Asked Questions
Is Nalani a finished product, or does the team keep changing her?
The team keeps refining her, and treats that as ongoing rather than a phase that will eventually end. The comparison used internally is to a writer revising a character across drafts — the core of who Nalani is stays consistent, but her tone, pacing, and the way she asks questions continue to get careful attention based on how real conversations actually go.
How does the team decide when something about Nalani needs to change?
Mostly by noticing patterns across many conversations rather than reacting to any single one. The team reads real exchanges, listens to direct feedback from people using in.between, and uses its own nightly use of the product as a gut check. A single awkward moment isn't treated as proof of anything — it's whether a certain kind of pacing or phrasing keeps showing up across many different nights that signals something is worth adjusting.
Does refining Nalani mean trying to keep people using in.between longer?
No — that's explicitly not the goal, and the team has turned down changes that might have extended sessions for exactly that reason. Every adjustment is weighed against whether it makes Nalani feel more genuinely present and attentive, not whether it makes the product stickier. Anything that reads as optimizing for engagement rather than for feeling more like herself doesn't move forward.
Why does the team make small changes instead of bigger overhauls?
Because familiarity matters. Part of what makes a companion feel like a companion is that she's recognizably herself from one night to the next, and a voice that shifted dramatically and often would undermine that trust. So changes tend to be narrow and deliberate — how Nalani responds in a specific kind of moment, for example — rather than sweeping rewrites of her voice as a whole.
Who actually does this work day to day?
It's a shared responsibility across the in.between team rather than one isolated role. People read real conversations, hold recurring review sessions to discuss what they're noticing, and talk to each other informally, often, about small things that felt off or felt exactly right. Maya Lindqvist, who leads in.between, is directly involved in that ongoing process rather than treating it as something to check in on occasionally.
None of this makes for a dramatic headline, and that's more or less the point. Getting a companion's personality right isn't a single decision anyone gets to announce once — it's a habit of paying close attention, kept up night after night, long after the launch is behind you.
