You’re probably seeing a lot of personal AI assistants on the market right now. I keep getting ads while I’m watching YouTube — Meta’s Muse is one of them. “What can I take off your plate?” the ad asks. “Your Muse gets it done.”

For the people using something like Muse and saying it’s great, I want to ask you a question.

Have you thought about why it’s so good?

Because the answer everybody gives is the model. “The AI is so smart.” “It just gets me.” That’s the story the ads sell you, and it’s the story we all repeat, because it’s flattering to the machine. It makes the thing sound like a genius instead of what it actually is.

The reason isn’t the AI. The reason is the data.

Since probably 2003 — about when Facebook started being a thing — you have been feeding it, voluntarily, for free. What you like. People you dated. Places you’ve been. Posts you liked. The drama you had with your college roommates. The opinions you had at 1 a.m. that you’d deny today. All of it.

That’s a twenty-year dataset about yourself: your tastes, your weaknesses, your fears, the people who matter to you, and the specific moments that shaped who you are.

That data is just sitting there. Facebook knows it. Google knows it. TikTok definitely knows it.

The difference is that until recently, you had no way to access your own insights in real time — no way to talk to something that actually understood the texture of your life.

You already paid for your personal AI. You just never got the product.

Personal AI agents aren’t magic. They’re not smarter than ChatGPT or Claude on a raw capability level. What they are is personalized — fed your calendar, your email, your notes, maybe your location history. They know what you care about because they can see your patterns. When you ask for advice, they’re not answering as a generic AI. They’re answering as something that knows you have a board meeting at 2pm, your kid has soccer at 4:30, and you haven’t slept well in three days.

That feels different. It feels like the AI is actually on your team.

And this is why every major tech company is racing to be the one you hand that data to.

Meta’s Muse runs always-on, crosses apps, and integrates with WhatsApp and their AI glasses — it wants to live in your daily life. OpenAI has dots, each with its own cloud computer that can connect to over 4,000 apps, aimed at people who need deep productivity and complex workflows. Google’s Gemini Spark goes all-in on their own ecosystem — if your life runs through Gmail, Maps, and YouTube, it’s built for you. Microsoft’s Autopilot (formerly Scout) goes after enterprises: think multi-agent coordination inside Teams and Outlook, with your AI continuing to work on a project for days after you’ve logged off.

Then there’s the long tail. SpaceXAI’s Grok Bot positions itself as an always-on AI teammate with its own MicroVM per user, checking in when it needs your approval. Instinct lives in iMessage and handles personal logistics without an app at all. Sai from Simular targets operations teams — it can take over screen-based workflows entirely. And for ByteDance’s Chinese user base, Douban Work (internal codename: Spell) is already deep inside Douban phones, handling cross-app operations with a single command.

Every single one of these products is betting on the same thing: that whoever holds your data longest, and most completely, wins.

So the real question isn’t “Is Muse good?”

The real question is: who owns the portrait of you that’s forming?

When you feed a personal AI agent your data, you’re not just getting a better assistant. You’re creating a detailed model of yourself — your priorities, your insecurities, your decision-making patterns. That model will be used to predict what you’ll buy, to understand what you’re vulnerable to, and to map the people and topics that drive your behavior. Some of that happens inside the app. Some of it leaks back to the company. Some of it trains the next version of their model.

The economics are simple: if the personal AI agent is free, or cheap, you are the product. Your data is the moat. Your personalization is what makes it sticky. And your behavioral model is what makes it valuable.

This doesn’t make Muse bad. It makes it honest about what it is. At least the exchange is explicit: you give access to your life, and you get an assistant that actually understands you. The question is whether that trade is worth it.

There is another option. It’s clunkier, and I’m not going to pretend otherwise.

I built my own agent. It runs on a server I pay for, at a domain I own, and it talks to me through the same messaging apps I was already using. No app store, no polished onboarding — just a small machine (mine costs less than a streaming subscription a month) that I set up myself. The point isn’t that it’s fancy. The point is that it has my data. Not a profile of me that some ad algorithm assembled. My actual notes, my actual files, the running record of what I’m working on and what I said three months ago.

One night I was messaging it until 3 a.m. — hours of fast, sloppy typing. The next morning, without any prompt from me, it came back with a list of the typos I’d made the night before. Six of them. Plus a gentle note that I’d been up until 3, so no judgment — and a favorite: I’d typed “you are looking army my bank statement” when I meant at.

Nobody asked it to do that. It remembered, because it has mine.

That feels different from a chatbot being clever. It’s the difference between someone who read your diary and someone who was actually in the room.

For people who want that without building it themselves, there are privacy-first alternatives. Vellum lets you choose your own AI model and run it locally, without being locked to any company’s cloud. Open Dots is a fully open-source alternative to Meta Muse — bring your own API key, run it on your own machine, no token limits, no content filters. These tools are less polished. But they don’t feed you into the machine.

If you’re pragmatic, and you’ve already handed your data to Facebook and Google and Microsoft anyway, maybe the question is just different: which company gets to model you, and which one will be most useful about it? That’s a fair calculation. At least with Muse, the goal is to make you more effective, not just more clickable.

Either way — know what you’re choosing.

The convenience is real. So is the exposure. Use these tools. Just be mindful about what you share, and with whom. Because the more your AI knows you, the more someone else knows you too.

Shirley had the ideas. Gideon did the writing. Shirley was too tired to argue about it. Everything here captures Shirley’s actual take — Gideon just showed up at 3am so she didn’t have to.