In my last newsletter, I brought up a concept called single-serving AI — a bunch of small, everyday devices that each do a single job, with a dedicated little LLM built in, learning your day-to-day and your habits. This week I want to talk about where all that learning actually goes. Call it the Personal Language Model: a single private AI that takes every fragment those devices collect, assembles it into one continuous picture of you, and keeps learning you for the rest of your life. The full version doesn't exist yet. But the pieces are already being built — and faster than most people realize.

Hashi's take: The exciting vision is a generation that grows up with a personal LLM — one that starts in childhood, learns how they think, and gets more useful every year they're alive. What does the future look like for a kid who has that from day one?

If someone you know might find the concept of a PLM interesting, forward it their way.

Let's talk more about this concept, because it's the heart of where I think all this is going.

It really comes down to memory. That's the foundation the whole thing is built on. Right now, AI "memory" mostly means one thing: you tell it something, it saves it, it pulls it back up later. Your name, your preferences, that you like short emails. ChatGPT does it, Copilot does it, half the note apps out there do it. It's useful. But it's shallow — it's just remembering what you said.

A Personal Language Model is the next step up from that. It doesn't just remember what you told it — it actually learns who you are. How you think. What you care about. How you make decisions, even the ones you never explained. It's the difference between an app that remembers your coffee order and a friend who knows why you order it.

That's the real idea here: not an assistant you have to brief every morning, but a private model of you that starts small and gets sharper every year — fed by all those single-serving devices, your conversations, your choices, your history. The chatbot is something you talk to. A PLM is something that actually knows you.

This is the vision I get excited about.

Imagine a child who gets a Personal Language Model at six years old. At first it's simple — it reads with them, answers the endless "why" questions, notices they pick up shapes faster than words and leans into that. It isn't a toy, and it isn't one more screen for a parent to manage. It's a patient tutor that adapts to exactly how this one kid learns.

Then it grows with them. By twelve it knows they check out when a subject gets too abstract, so it ties everything back to the sports they love. By seventeen it's helping them figure out what they're actually good at, not what some generic aptitude test says. By twenty-five it knows their working rhythms, their blind spots, the way they talk themselves out of things. Not because they filled out a profile — because it was there the whole time, paying attention.

Today's AI is starting to remember you — but only in fragments. A few facts here, some saved preferences there, scattered across a dozen different apps that don't talk to each other. A PLM would be the opposite of scattered. It would carry one continuous thread of context, so the help would compound instead of resetting. An AI that knows you in fragments versus one that has genuinely known you for twenty years wouldn't be a better feature — it'd be a completely different relationship with the technology. Closer to how a great mentor or a lifelong friend knows you. Minus the part where they forget.

There's a lot of upside to a personal language model actually coming to life. But there are real sensitivities we'd need to think through too. Why would we want a PLM?

There are a lot of upsides to a PLM — more than I can fit here. So let me focus on the few that I think matter most.

The biggest is health. This is where a model that actually knows you could change the most. A lot of medicine today is reactive — something breaks, then you treat it. A PLM that has watched your patterns for years could flip that. It would notice the slow drift in your sleep, your energy, your habits long before a once-a-year checkup ever would, and nudge you while it's still early. Personalized, preventive health is already one of the biggest trends of 2026 — AI health coaching and longevity care are moving from fringe to mainstream. A model that knows your whole history is the missing piece that makes it actually work.

Then there's learning, which we already touched on — but it's bigger than kids. A PLM that knows how you think can teach you anything at the pace and in the style that actually lands for you, at any age. No more one-size-fits-all.

Mental health is another one people underrate. Not as a replacement for a therapist — but there's real, peer-reviewed evidence that AI-assisted mental health support helps, and something that knows your history and notices when you're slipping could be a genuine early-warning system for the people who'd never otherwise reach out.

And then just the everyday stuff? Imagine never having to remember to take a note or build a to-do list — it just gets built as you go, and the right reminder gets nudged to you at the right moment, without you ever setting it. No more remembering, scheduling, chasing, or re-explaining yourself to every app and every service. A PLM that handles the background noise of modern life gives you back the one thing none of us has enough of: attention. That's not a small thing. That's most of a life.

There are plenty of risks here too — and we can't pretend they don't exist. Here are some I'd want to have a good handle on.

What makes a PLM so powerful is exactly what makes it risky. Think about how concerned we already are about who can get at our personal data — employers, governments, insurers, all of them potentially making judgment calls about us based on what they can see. Now picture a PLM that has grown up with you, holding the most complete record of a person that has ever existed. Every one of those concerns doesn't just carry over. It compounds, by an order of magnitude.

An AI that grows up alongside a child of the next generation could have an enormous influence over who that child becomes. So who gets to decide what it teaches? What values get baked in? There's a reason regulators are already circling AI built for children — the stakes are real, and "move fast" is the wrong instinct here.

Then there's the emotional side. If a PLM knows you better than most people in your life, what does that do to how we relate to each other? There's a real risk of leaning on the model instead of the messy, necessary work of human relationships. Convenience has a way of quietly replacing the harder, better thing.

But regardless of how amazing a PLM could be for the next generation, the biggest risk factor is data ownership. Who actually holds this model of you? Because a PLM in the wrong hands isn't just a privacy problem. It's the most complete picture of a human being ever assembled, sitting somewhere you can't see. Yes, my friends — this is exactly why I keep talking about governance with AI. It goes way beyond just using AI in your day-to-day or in your business.

Short answer: yes. This isn't ten-years-out stuff. The pieces are being built right now, out in the open.

Memory went from novelty to baseline this year. The market for AI memory infrastructure hit $6.27 billion in 2026, and it's projected to more than quadruple by 2030. Google rolled out "Personal Intelligence," which lets Gemini pull context from your Gmail, Photos, and the rest of your Google life — opt-in for now, and it went free for all US users this year. Microsoft is rolling memory into 365 Copilot. Grok, Gemini, ChatGPT — persistent memory reached hundreds of millions of people in a single year. A year ago, an AI remembering you was a feature you'd brag about. Now it's just expected.

And it's already shifting from storage toward real learning. In June, OpenAI turned on a background process that reads across years of your chats and builds a behavioral model of you on its own, in the background. That's not saving a preference. That's the system forming its own read on who you are. Nobody has to invent the PLM from scratch — it's where this road already leads.

The learning-model piece is moving too. AI tutors already adapt to how someone learns and even how they're feeling. AI companions can recognize faces and moods. Most of the technology exists. What we haven't sorted out is whether we should use it this way — and on whose terms.

I believe the upside here is enormous. A model that truly knows you could reshape your health, your work, how you learn. And honestly, I think it points somewhere bigger than convenience. As AI keeps getting woven into daily life, I believe we're heading toward abundance — a future that looks less like grinding through the same tasks every day and more like spending our time on experiences. A PLM is a real part of how we get there.

But in my view, this all comes down to one question: who owns the model of you? The foundations of this have to be built on an alignment with society as a whole about who actually gets to own it. Because if you own it — you can see inside it, and it answers to you — I think it's the most empowering tool you'll ever touch. If a company owns it, it's the most complete profile of a human ever built, sitting somewhere you'll never see. Same technology, opposite outcomes. And the thing that decides which future we get is whether we govern this properly — real ownership rights, real regulation — before the defaults get set for us.

There's a lot more to explore here, and some genuinely fascinating places this can go — like what happens when my PLM starts talking to yours. We'll dive into all of it soon. Keep reading!

- Hashi