You carry a rich, nuanced understanding of every important person in your life: their preferences, their communication style, what they care about, what annoys them, what they need from you. Your AI agent has none of this. It knows you but it doesn't know anyone you know, which means it can't draft a message in the right tone, suggest a gift worth giving, or prep you for a conversation with context about the other person.
Michael Tiffany
Your agent can plan a dinner that respects your dietary constraints, but it can't plan a dinner party unless it knows that your sister is vegan, your college friend has a severe shellfish allergy, and your neighbor will eat anything but wants his wine paired. It can prep you for a meeting, but the brief is hollow if it doesn't know the person on the other end. It can suggest a gift, but research on giver-recipient mismatches consistently shows that givers overweight relationship signaling and underweight recipient preferences, and a Yale School of Management review found that this gap nearly disappeared when givers focused on what they actually knew about the recipient rather than on what they wanted the gift to say about themselves. In other words, what makes a gift feel personal is evidence that the giver knows the recipient, which requires exactly the kind of knowledge your agent currently lacks.
I think this is the most foundational gap in most people's AI setup, and the most under-appreciated. Robin Dunbar's research on cognitive limits to social relationships suggests that humans can maintain roughly 150 stable relationships, layered into concentric circles of increasing distance: about 5 intimate relationships, 15 close friends, 50 casual friends, and 150 acquaintances. A 2021 Royal Society review challenged the precision of the 150 figure but confirmed the underlying principle: there is a real cognitive constraint on how many relationships we can track, and the constraint operates through the same limited resource that makes all the other domains in this series hard to manage manually: memory and attention.
The temptation is to treat this like a contact database: name, relationship, birthday, phone number. That's the information on a Rolodex, and it's useful for reminders but worthless for judgment. Knowing that your brother's birthday is March 14th tells your agent when to remind you, but it tells the agent nothing about what to give him, how to word the message, or whether he'd prefer you call instead of text.
The knowledge that matters is relational and operational: how this person relates to you, what they care about, how they communicate, what they need from you, and what your history together looks like. I think of it as the answer to a specific question: if a mutual friend asked you to describe this person so they could interact with them competently, what would you say?
"My brother is a high school science teacher in Portland. He's quiet and thoughtful in conversation and hates being put on the spot. He reads constantly, mostly history and science fiction, and he's particular enough about books that a gift card is safer than a specific title unless you're certain. He's terrible at responding to texts but will always pick up a phone call. We talk about once a month, and he's closer to our mom than I am, so he usually has family news I haven't heard yet. His wife's name is Elena, their daughter just started kindergarten, and they're renovating their kitchen, which has been a source of stress for six months."
That paragraph contains a communication preference (call, don't text), a gift-giving constraint (gift cards over specific books unless certain), a relationship dynamic (closer to mom, has family news first), a current life context (kitchen renovation stress), and enough specificity about interests that an AI could reason about what would make a good birthday gift.
The personal CRM movement has been trying to solve this problem for years, and I think most implementations fail because they treat relationship management as data entry: fill in the fields, tag the contacts, set the reminders. That approach optimizes for completeness at the expense of the thing that actually makes relationship knowledge useful, which is richness. Ten fully tagged contacts with name-birthday-company are less useful than two contacts described the way you'd describe them to a friend.
So don't start with a system; start with five people. Pick the five people you interact with most frequently or most consequentially: your partner, a close friend, your boss, a parent, a sibling. For each one, spend two minutes telling your agent about them the way you'd brief a mutual friend. Cover how they communicate, what they care about right now, what your relationship dynamic looks like, and any context that would help someone interact with them on your behalf.
You're not building a database. The depth matters more than the breadth. Once your agent genuinely understands these five people, you can extend the same approach to others as the need arises: when a new colleague becomes important, when a friendship deepens, when a service provider becomes a regular part of your life.
Your agent can draft messages calibrated to the recipient. It knows that your brother prefers phone calls, that your boss likes concise emails with action items at the top, and that your college friend communicates almost entirely through voice memos. It can suggest gifts that demonstrate genuine knowledge: not "books for men aged 35-45" but "your brother reads history and sci-fi, is renovating his kitchen, and prefers practical gifts." The meetings article can teach your agent to prep you with context about the person you're meeting; but with relationship knowledge, those briefs gain a human dimension that calendar data alone can't provide. And it can help you coordinate care for kids or pets by knowing which grandparent is best with the toddler or which neighbor your dog actually likes.
The ongoing maintenance is the same pattern as several other articles in this series: narrate after interactions, and only when something worth remembering happens.
"Had lunch with David. He's leaving his job at the end of the month, hasn't told many people yet. He's looking at startups and asked if I knew anyone hiring. His daughter got into the college she wanted, which he's thrilled about."
That post-interaction note takes thirty seconds and gives your agent four updates: a career change, a confidential status, a request for help you might act on, and a family milestone.
You don't need to debrief after every interaction. You need to debrief after the ones that change something: a life event, a new piece of context, a shift in the relationship, a request you want to remember. Roberts and Dunbar (2015) found that relationships decay predictably without active maintenance, and that the effort required to prevent decay varies by relationship type: friendships degrade faster than family ties and require more frequent contact to sustain. Your agent can't replace that contact, but it can make sure you show up to each interaction with the context to make it count.
Storing detailed information about other people carries ethical weight that storing information about yourself does not. The people in your life haven't consented to being profiled by your AI agent, and some of the most useful relationship knowledge is also the most sensitive.
I think the right approach is to store what you'd be comfortable sharing with the person if they asked. "I told my AI that you prefer phone calls over texts" is something most people would find thoughtful. "I told my AI about your marital problems" is something most people would find invasive. The line is contextual and personal, and you should think about it for each relationship rather than applying a blanket policy. Your AI tool's privacy policy matters here too; understand where the data is stored, whether it's used for training, and whether you can delete it.
How many people should I teach my agent about? Start with five and expand as needed. Dunbar's hierarchy suggests we maintain about 5 intimate relationships and 15 close ones; those inner circles are where the deepest knowledge lives and where your agent's help is most valuable. You'll naturally extend to others when specific situations demand it.
Should I include professional contacts? If you interact with them regularly, yes. The meetings article already established the value of pre-meeting briefs, relationship knowledge gives those briefs a human dimension that a calendar invite alone can't provide.
What if a relationship changes or ends? Note the change: "David and I had a falling out. I don't want to reach out right now." Your agent should respect that boundary and stop surfacing the person in suggestions until you say otherwise.
How does this connect to the gift-giving article later in the series? The gift-giving article depends almost entirely on this one. A good gift requires knowing what someone cares about, what they already own, and what would surprise them. Without relationship knowledge, your agent is guessing based on demographics, which is how people end up giving generic candles.
Copy and paste the prompt below into your AI agent to get started. Then, pick your five, describe them, and see if the gift suggestion and the message draft feel like they came from someone who knows the person. The gap between what your agent produces and what you'd actually send is your teaching material for the next round.
I'm going to teach you about the important people in my life so you can help me with gifts, messages, meeting prep, and coordination. For each person, I'll describe who they are, how they communicate, what they care about right now, and what our relationship dynamic looks like. Let's start with five people I interact with most. After I describe them, suggest a birthday gift for the first person I mention and draft a short check-in message for the second. I'll tell you how close you got.
Fulcra was designed by people who get privacy and know the importance of an infrastructure solution that can be the secure private datastore for the rest of your life. Here data is yours, under your control, and only shared with the people and tools you choose to share it with.
