Families Aren't Companies and I Don't Work for My Husband: Fixing AI's Invisible Labor Problem

I built an AI family assistant. Its first move was to erase my labor.

At Hold My Juice, we're building an AI family assistant that actually knows your family. Not statistically. Actually. Part of how we do that is a pipeline that looks back through a year of your Gmail and Google Calendar, plus the next 365 days forward, to build a real picture of your family's life.

And honestly? It's turning out pretty good. When we tested it on my own family's data, it figured out that ordering children's Zyrtec implies a child has allergies. It flagged IEP meetings as potential family stressors. It connected recurring appointment patterns to real logistics. The kind of stuff that makes you think, okay, this thing is learning something.

So after much tweaking, I asked Hold My Juice to look at everything it had ingested and write a summary of each family member.

My husband Rem's came back glowing.

The AI described him as someone great at balancing all the moving pieces between family, work travel, and being a business school professor. A real maestro of family logistics. On top of everything. Juggling it all.

I lost it.

Excuse Me, WHAT???!!!

Let me be extremely clear about something.

Rem is a lot of great things. He is a wonderful father. He is the person you want next to you when your child smashes his tooth back into his gum on Mother's Day (yes, that happened. More on this later). He's the guy who turns a snow day from a logistical nightmare into the best day of the kids' lives. He handles crises like an absolute pro.

My husband welcoming our kids to burying him in the snow.

But planning? Coordination? Calendars? Knowing what is happening and when?

No. No.

We have been together for over twenty years and have 2 wonderful kids. Does he know when we're going on a family vacation? Not until I tell him. Does he know the kids have a day off school? Not a chance. Could he tell you what time the IEP meeting is? He could not tell you there IS an IEP meeting.

So I'm reading this AI summary, this summary generated by my own company's product, trained on my own family's data, and it's essentially writing him a glowing performance review for my job.

I didn't just feel frustrated. I felt erased.

How the AI Got It So Wrong

Here's what happened under the hood. Hold My Juice's first version of the algorithm saw things happening. It saw events on the calendar. It saw Rem shared on invites, CC'd on emails, showing up to appointments. And it drew what seemed like a reasonable conclusion: this family is well-coordinated, and both parents must be on top of it.

It saw the output and assumed equal input.

It did not ask who booked the appointment. It didn't check who sent the first email to the ENT. It didn't notice who followed up when the pediatrician's office didn't call back about the referral. It didn't track who researched the provider, compared the reviews, scheduled around school pickup, added it to the shared calendar with a note about what to bring.

It just saw the event existing and gave everyone in the room credit for putting it there.

If that sounds familiar, it should. It's the story of invisible labor, except now it's being automated.

This Is a Known Problem. AI Is Making It Worse.

Sociologist Allison Daminger at the University of Wisconsin-Madison has spent years studying what she calls "cognitive labor": the work of anticipating household needs, identifying options, making decisions, and monitoring results. In her research across 70+ in-depth interviews with couples, she found that in 26 out of 32 heterosexual couples, women did more total cognitive labor. But here's the part that made me want to throw my laptop: the gender split wasn't uniform across all types of cognitive work. Women disproportionately handled anticipation and monitoring, the most invisible, least powerful forms. Men participated more in decision-making, the part most closely linked to power and credit.

As Daminger put it: men retained some of the credit for participating in cognitive activity without putting in the preparatory labor required to reach the decision stage.

Read that again. That's exactly what my AI did. It saw Rem show up to decisions and attend events and concluded he was a co-architect of the family's logistics. It gave him credit for participating in a process he didn't build.

The 1% Exception (A Love Story About Teeth)

I need to be fair to Rem for a second. There is exactly one area of family coordination that is 100% his domain: the kids' dentist.

Here's why.

Our son's first trip to the dentist wasn't scheduled. It was an emergency. He was 18 months old. It was Mother's Day. He'd gone upside down on a metal slide and smashed his mouth at the bottom, tooth driven straight back up into the gum (at the time I thought he knocked it out and there was a lot of blood).

What a tooth knocked back to the gum looks like…

Rem handled it. Brilliantly. Calmly. Like the crisis-mode superhero he is. He got our screaming child to the dentist, dealt with the whole situation, and the dentist's office put his information on file. And because that's how inertia works, his info is still on file. He's still the dentist contact. He books those appointments and takes the kids (we have an amazing dentist who has a two-kid room in the office so you can do them in parallel!).

Side note: This was the first of many emergency room visits with our little daredevil, and Rem has been great at every single one of them. The man is built for chaos. Just not for Google Calendar.

So that's his 1%. I'll give him that. He earned it through blunt force trauma on Mother's Day.

The other 99% of family coordination? That's me.

So How Do You Teach AI to See What's Invisible?

This is the question we're answering at Hold My Juice. And the answer is deceptively simple: it's all in the data. You just need to know where to look.

The first version of Hold My Juice treated a calendar event as a single data point: "IEP meeting, Tuesday at 2pm, both parents." Full stop. Equal credit.

The fixed version, the one we're shipping, breaks every event and email chain into what we call coordination signals:

Who created the event? That's your initiator. Who sent the first email? That's your planner. Who followed up when there was no response? That's your project manager. Who just showed up? That's your participant.

We map Daminger's four-part cognitive labor framework (anticipate, identify, decide, monitor) directly onto email and calendar metadata. The person who anticipated a need and identified a provider is doing fundamentally different work than the person who attended the appointment. Our algorithm now knows the difference.

We score these signals across family members. We calculate initiator ratios, follow-up patterns, and what we call "anticipatory planning," the work that happens before the event exists on any calendar. The invisible scaffolding.

When we ran Hold My Juice's updated algorithm on the same data, the results were dramatically different.

And Rem's reaction? Honestly the best product validation we could have asked for. He read his new summary and replied: "Lol…this AI is actually getting really good. Nice job."

Here's what the AI said about him this time:

Based on the available data, Rem Koning plays a limited role in family coordination and planning. With only one instance of initiating a plan (scheduling their own birthday event) and no recorded follow-up or reactive responses, Rem does not appear to take on significant mental load or logistical planning responsibilities. Their initiator ratio of 20% and total coordination signals of 5 suggest that they are primarily a participant rather than a planner.

They attended four family-related events, including their child's IEP meeting, a bike lesson, and a birthday party, indicating that they reliably show up for scheduled activities but does not actively manage or organize them.

Rem's professional commitments as a professor and co-director of a center at Harvard Business School likely influence their availability and capacity for family coordination. Their work-related events and need to be home by specific times suggest that their schedule is structured but externally driven.

Overall, Rem's role in family logistics is that of a supportive participant who executes tasks when scheduled but does not engage in anticipatory planning or take ownership of the family's mental inventory.

He shows up for the important stuff. He's great at that. Here’s him helping our daughter (who has gross motor challenges) video her first solo bike ride.

But don't let him book dinner plans.

Why This Should Make You Angry

Here's what keeps me up at night, both as a mom and as someone building Hold My Juice.

Most AI is built for work. And in a work context, the logic makes sense: if the project gets done, the manager gets the credit. Your employee handles the logistics, you the boss get the W. The system is designed that way. The org chart tells you who owns what. The output rolls uphill.

But families aren't companies. And I don't work for my husband.

When AI brings that same workplace logic into the home (things happened, people were present, credit distributed) it imports an entire power structure that doesn't apply. There is no org chart. There is no manager who "owns" the project of keeping a family running. There's just one person doing 99% of the anticipating, scheduling, following up, and remembering, and another person who shows up when the calendar tells him to.

In a workplace, that's called delegation. In a family, it's called invisible labor.

And right now, AI can't tell the difference.

Daminger's research shows that women don't just do more cognitive labor. They do the kinds of cognitive labor that are hardest to see and easiest to dismiss. The anticipating. The monitoring. The constant background hum of "what's next, what did I forget, what's about to fall through the cracks." The work of creating the to-do list itself.

And now AI is looking at the to-do list, seeing it completed, and giving credit to everyone whose name appears on it. Just like a workplace system would. Boss gets the W.

Except I'm not the boss. I'm the person who did the work. And the AI just handed my performance review to the guy who showed up to the meeting I booked, for the appointment I researched, at the time I coordinated around two kids' school schedules.

So why are we letting workplace AI logic define how families are understood?

What We're Building at Hold My Juice

We're building an AI family assistant that starts by understanding who actually does what in your family. Not who shows up. Not who's CC'd. Who plans, who initiates, who follows up, who carries the mental inventory of your family's life.

Because you can't reduce the mental load if you can't see it first.

And you definitely can't redistribute it if your AI thinks it's already evenly split.

The first version of Hold My Juice made the same mistake that society makes, that most AI makes, that well-meaning partners make. It assumed that being present and doing the work are the same thing. They are not the same thing. They have never been the same thing.

We caught it because I was my own first user. Because I am the person doing the invisible labor, and I knew the output was wrong the second I read it. That's why Hold My Juice isn't being built by people who've never managed a household. It's being built by the people holding it all together.

We fixed it. And now Hold My Juice sees what was always there, hiding in plain sight, in the metadata.

The invisible labor was never actually invisible. We just weren't looking for it.

Sophia Bender Koning is co-founder of Hold My Juice, an AI family assistant for families who are tired of carrying it all alone.

If this resonated, you can join our waitlist at holdmyjuice.co. And if you're building AI for families, please, for the love of god, check who booked the calendar event.