The People in Your Notes
Keeping structured records about people isn't new. What changes when AI can read them is subtler than it looks.
When employees left companies in the 1980s with their Rolodexes, they sometimes got sued for it. The legal argument was straightforward: the contact list had been assembled on company time, using company resources, in service of company relationships. The names, the notes in the margins, the handwritten preferences next to each card — all of it had commercial value. Courts had to decide whether a personal network could belong to an employer.
They mostly found that it couldn't. The contact list reflected personal judgment — which relationships mattered, what was worth remembering, how to characterize someone — and judgment that intimate can't be cleanly separated from the person who exercised it. The Rolodex was invented by Hildaur Neilsen in 1956. By the 1980s it was already legally contested property. The question of who owns knowledge about relationships turns out to be old and unresolved.
It's returning now, in a different form.
The note is not the fact
People in your life tell you things. Some of it you need to act on. Some of it makes you better at caring for them. A friend mentions in passing that her mother has been unwell. A colleague, over lunch, tells you he's been quietly looking at other roles. A contact's preference for walking meetings over sitting ones. You write these down not because you're building a dossier, but because you'll forget, and forgetting feels like failing to pay attention.
Personal relationship tools — Monica, Clay, Dex, the Notion templates that get reshared every six months — are built on exactly this premise. Monica's GitHub description is "remember everything about your friends, family and business relationships." It has been starred over 20,000 times. The behavior is clearly wanted.
What's new isn't the behavior. People have kept written records about other people for as long as writing existed. What's new is machine-readability, and what that changes is subtle enough to miss.
When you write "Sarah: prefers direct feedback, mentioned a difficult family situation" into a structured person record, you're doing something that looks like recording a fact. It reads like a fact when you look at it later. But it's a compression of something more complicated: a moment, a conversation with a particular tone, a thing she said that you interpreted through your own frame, at a specific point in time.
That note goes through your head before you use it. You remember that the family situation was months ago and she seems fine now. You remember that "prefers direct feedback" was her description of herself in a specific context, not a standing property. You apply social judgment automatically, because you know her.
An AI reading that record doesn't have that filter. It sees two fields of equal weight, equally current, equally authoritative. The divorce might be years behind her. The preference might have been your projection. But the record doesn't know this — and neither does the system reading it.
This is the design problem that most personal CRM tools haven't addressed. They optimize for adding information but not for representing what kind of information it is. The source. The certainty. Whether something is still true. A person record with ten outdated fields is worse than one with two current ones, but there's no visible difference unless you already know the person well enough not to need the record at all.
The value of actually remembering
None of this argues against keeping person records. The case for them is real: you have a meaningful conversation with someone at a conference and want to follow up six months later, but you can't remember the specifics. You mentored someone two years ago and they reach out again — and you want to remember who they were, what mattered to them, what you actually said. A person record is how you do that without relying on a memory that's already carrying too much.
Robin Dunbar's research on social cognition puts the limit of unaided relationship maintenance at around 150 people — not because humans are incapable of knowing more, but because the cognitive overhead of tracking context, history, and preference eventually wins. External records push that ceiling up, in the same way that writing things down lets you think about more things at once. The question is whether the records supplement knowing someone or substitute for it.
Good use looks like this: you check a person record before reaching out, get reoriented, and then actually talk to them. The record briefed you. You're still the one in the room.
Riskier use looks like this: an AI queries the records on your behalf and acts on what it finds — decides what tone to use, what to reference, what to avoid — without your judgment in the loop, without the context you hold that you didn't write down, without any sense of what has changed since you last updated the entry.
What local actually means
Person records in Harbor live locally. They're scoped — an AI tool can read them only if you've explicitly permitted it to, and you can revoke that permission. They don't get synced to a cloud service that might be breached, indexed for training, or accessible to something you didn't authorize.
That's not a compliance decision. The GDPR has a household exemption that covers personal notes about people in your life, and most jurisdictions haven't caught up to machine-readable personal records anyway. It's a design position about who the data serves.
A person record kept locally is the digital equivalent of a handwritten card in a Rolodex: yours, organized by your judgment, visible only to what you've permitted to read it. The Rolodex lawsuits mostly failed because courts found that personal judgment can't be cleanly transferred. The same logic applies. The relationships are yours. The notes are yours.
What you've chosen to remember about someone says something about you, not just about them. That's worth keeping close.
Asgeir Albretsen is the founder of Harbor.