There are two very different futures being built with AI right now.

One is centralized, filtered, optimized for scale, and quietly paternalistic.

It decides which questions are acceptable, which answers are safe, and which ideas never quite make it to the surface.

The other is quiet, local, intimate, private, and far more human.

I’m choosing the second.

Today, I want to talk about why I’m building my own personal AI…

How you can do the same—and why it’s easier than it looks…

Why doing it right now matters…

And why personal AI feels like the next personal computer revolution.

The Wrong Fear

Most public conversations about AI fixate on scale: superintelligence(!), regulation(!), alignment(!), existential risk(!).

Some of those questions are real. But they’re also downstream. And mostly irrelevant.

The more fundamental issue is ownership.

If intelligence exists only inside centralized systems—cloud models trained, filtered, and constrained by a small number of institutions—human agency shrinks by default.

And that conclusion doesn’t require bad actors. It holds even if the people running these systems are (and remain) good-intentioned angels.

Human agency shrinks because of one thing: it’ll be convenient.

Nobody wants a future where their grandchildren inherit a single sanctioned version of reality, delivered through a “safe” interface that decides which questions are allowed to be asked.

Most want a future where technology makes them more resilient and makes their lives better and more interesting…

Where it helps them do more with less, move through life with more engaging problems, and meet the world with curiosity intact.

That’s personal AI.

Why “Personal” Means Local

Right now, most AI lives somewhere else—on servers you don’t own, governed by policies you didn’t choose, changing without your consent.

One update can alter its behavior. One terms-of-service change can reshape what’s allowed.

Running AI locally shifts that balance. No subscriptions. No rate limits. No silent policy changes.

The model lives on your machine, runs when you want, and stays exactly as you leave it.

Your data never leaves your computer. Prompts aren’t logged somewhere else. Context isn’t upstreamed or monetized. Privacy isn’t a “feature”—it’s the default.

This matters once AI begins accumulating context: your work, your writing, your patterns, your unfinished thoughts. That context belongs with you.

A local AI has no ambient access. It can’t wander your files or observe in the background without your permission. Nothing uploads anywhere. Nothing persists elsewhere. You decide what’s visible, what’s indexed, and what disappears.

And right now, it’s more possible than ever.

The Old Tradeoff Is Gone

Local AI used to mean compromise. Slow models. Crude interfaces. Limited capability.

BUT…

That’s no longer true.

Open-source models have crossed a threshold. What once required a data center now runs on a laptop.

Training efficiency favors smaller models, and progress there is accelerating faster than in frontier systems.

Advances in quantization makes an AI model up to 70% smaller and easier to run, without ruining its usefulness. A model that once needed tens of gigabytes fits comfortably on consumer hardware. A single GPU—or a modern Mac with unified memory—goes a long way.

Meanwhile, the big labs are slowing down.

The largest gains are happening in the open, in models optimized for real machines and real users.

The innovation curve is bending toward people who run their own stack. And, increasingly, anyone can do it.

How Local AI Actually Works

It’s simpler than most people think.

An AI model is just a file—a large collection of learned weights.

To use it, you need three things:

1. The model file

2. An engine to run it

3. An interface to interact with it

Tools like Ollama handle all three.

Download a model. Run it locally. Expose it as an API so other programs on your computer can talk it the same way they talk to a printer driver, a database, or a media player.

Done.

From there, everything changes.

You can freeze versions. Fine-tune behavior. Shape memory. Run offline—on a plane, in a forest, during an outage.

New interfaces feel just as easy as cloud chat apps—but instead of hiding the machinery, they show it.

You can see how much the AI is thinking, how much memory it’s using, how much power it’s drawing, and which model is actually running. You’re no longer guessing. You can watch the system work in real time.

It feels less like software and more like a collaborator you understand.

The Next Personal Computer Moment

AI will shape how we think.

That’s no longer optional.

The real question is whether that thinking is mediated by centralized systems—or anchored in personal context.

I don’t want to outsource my memory, reflection, or meaning to someone’s data center.

I want a system that knows my history, understands my patterns, and helps me see myself clearly—without telling me who I’m supposed to be.

Escaping technology is not the play. Using it deliberately is.

I’ll write more about how I’m doing it soon.

Local AI matters because it shifts intelligence from something you rent into something you own.

And over time, intelligence naturally migrates to the edges—where learning accelerates, mistakes cost less, and people become more resilient by default.

More soon.