Every website is now a chatbot.
None of them know you. Each is trapped inside the company that built it, running a cheap model because they can’t afford to serve a good one to everyone.
What you want is one that knows you and can talk to all of them for you.
But who would it belong to?
You don’t own it
Right now it belongs to a handful of big companies, and the average person owns none of it. The labs keep their best models behind the top paid tiers and ration them with weekly limits.
Shut out, most people have soured on the whole thing. Depending on who you ask, AI means cheating, or slop, or surveillance, or a datacenter taking over their town. I used to think they were misinformed, and I blamed my Michigan friends for worrying about a datacenter draining their water supply. I was wrong. None of this was ever built for them to own.
And it can be taken from anyone. OpenAI pulled GPT-4o out from under the people who loved it. Tens of thousands signed petitions, the company brought it back for paying users, then retired it for good six months later. Anthropic shipped Fable 5, the best model anyone had released, and three days later the government ordered it pulled and the company shut it off for everyone. I finally realized how the crazy 4o people felt.
An alien intelligence
AI is the closest we have come to an alien mind arriving on earth. It does not think the way we do, live in a body, or learn the world through our senses, and at most of what we use our minds for it already matches us or beats us.
The first thing we did with something this powerful was shrink it into a chat box, which puts the work on you by assuming you can name exactly what you want in one prompt, and most of the time you cannot. A stranger ordering your lunch needs to know all the minute details to make the right choice, which a close friend, after years of observing you, just knows.
A chat box only works once the AI knows you that well. Then a vague prompt reaches the same place a detailed one would, because it fills in the rest from you.
You cannot front-load that with a hundred onboarding questions. No one finishes them, and the answers are thin. You learn a person by spending time with them, and a new AI learns your judgment the same slow way, making mistakes until you correct it.
I was inspired by Amy Deng, an AI researcher who spent about a month logging her health in obsessive detail, from her hourly energy to a hundred blood markers, then fed all of it to frontier models to chase down her chronic fatigue. She worked with her doctors the whole way, and together they pulled apart the causes. Her rule was that no amount of context is too much.
Knowing you is only half of it. The other half, acting for you, already exists in code. Cursor moved AI from the chat box into the editor. Codex takes a whole goal, runs it for days, and returns the finished work.
The unit of work keeps climbing toward a whole goal you hand off. You should not have to manage any of that. You say what you want and trust the right machinery to run.
Big Brother
These models are the closest thing we’ve had to an empathetic patient human, and all of it is controlled by a force beyond your power. Whoever owns it can switch it off whenever they choose, and they see everything you ever tell it.
People type the symptom they are too scared to google and the debt they are hiding from someone they love. We are more honest with a text box than with anyone we know, so it learns your inner life better than your closest friend does.
If you do not own it, every word goes to a company you are trusting to never look and never hand it over.
A company that holds your thoughts can hand them to a court or decide what you are allowed to ask.
Why it has to be yours
Personal AI must be owned by the person.
When you own it, there is nothing to subpoena and no one to report you. The record of your inner life stays with you. No company or government can rate limit or switch off a model you run yourself. Owning yours should not mean hoarding hardware or going off the grid. You should get to choose.
Computers did this first
Ownership has reached the person before. Computing started locked inside institutions. A mainframe filled a room, belonged to a university or a government, and you booked time on someone else’s machine. Then it moved out to the desk and into your pocket, and you went from borrowing time to carrying the machine. In 1977 Ken Olsen, who ran Digital Equipment, one of the largest computer companies in the world, said there was no reason for anyone to have a computer in their home.
AI is in that same mainframe era right now, locked in a few giant buildings you do not own. It will make the same move, from the institution to the person.
It should not be a luxury
Money is the only thing keeping this from everyone.
Most AI runs by renting one giant brain in the cloud for every task. Satya Nadella says there should be as many models in the world as there are firms, because a firm is a learning system. I would go further. As many models as there are people.
Open models are catching up fast. The open frontier, like GLM-5.2, trails the best closed models by only a few months, and smaller ones like Qwen3 and gpt-oss-20b already run on a laptop and handle most everyday tasks.
The richest people already delegate by talking, handing work to staff out loud.
OpenClaw, built by Peter Steinberger, runs beside you and touches the real tools on your machine, driven from the chat apps you already use. Garry Tan runs it with full context of his email and texts and calls it the better version.
It costs a fortune. Garry says he spends a million dollars a year on tokens, and Peter spends over a million a month. You can do the work of a thousand people right now, in Garry’s words. You just have to pay for it.
Even the affordable version runs about a thousand dollars a month, which puts it out of reach for almost everyone. The most capable intelligence ever built is already here, and it costs what a small staff costs.
The companies selling it are taking the loss. SemiAnalysis found that a $200 a month plan, used hard, pulls about $8,000 of compute on Claude and $14,000 on ChatGPT Pro, forty to seventy times what you pay. Anthropic had to cap Claude Code with weekly limits after one user burned tens of thousands of dollars of compute on a $200 plan. Even sold that far below cost, there is never enough, the labs perpetually out of GPUs and melting the ones they have. A middleman renting you a rationed model below cost, taking a cut of every token, cannot be how everyone ends up with one.
Almost all of that price is rent. The cloud charges you for hardware you will never own and stacks a margin on top, and the bill comes every month for as long as you use it. Renting one high-end chip there is about fifteen hundred dollars a month and never stops.
A machine of your own costs one to a few thousand dollars once, and after that only the electricity to run it.
It pays for itself within months, and from then on the intelligence is yours for the price of the power it draws.
People want infinite use at no cost. That arrives when compute gets cheap, and compute is falling by orders of magnitude. Mark Pincus, who built Zynga, told Garry that consumer is not investable right now, and that this is exactly the reason to build it. Pincus and Garry put the moment ordinary people get this at around 2029.
Anything that can be free will be free. The day a personal AI costs nothing, everyone has one.
What owning it looks like
An AI you own runs on your own hardware, all day for the cost of electricity. It does most of the work on the machine in front of you and reaches for the big cloud models only when it needs more. Companies are already moving real work onto local models and buying their own hardware to run them.
It watches your store or reads your email only because you put it on your own machine and pointed it there.
Right now that machine is a DGX Spark or a Mac Mini, a box you plug in under the desk and forget. A box you forget has no presence.
Interface
You talk to a thing like that. Every generation works a machine in a way the one before finds strange. The people before us grew up turning dials and pressing buttons that moved. We grew up on glass, tapping a flat screen that buzzes back to fake the click of a button, and we call that buzz touch.
The next one is the conversation. You say what you want and it happens, and the machine fades into the room. The chat boxes we started with were its first crude version.
The conversation has to reach you wherever you are, through the things you wear and carry. Today every wearable runs its own assistant that knows only a sliver of you, the same way every website runs its own chatbot. That context has to be unified, and your own AI is the thread that ties it together, the one continuous you that follows from your glasses to the machine at home. That is the argument for one company building all the wearables, so every one of them feeds the same mind and no company holds a separate piece of you.
People hear all this and jump to brain interfaces. I think those are a red herring for now. Virtual reality promised the same and never took off, and people stayed with their computers. A brain interface is the cleanest path from intent to action on paper. Getting one installed in your skull asks far too much of almost anyone for now.
Soon a model will handle nearly everything we use our minds for, and past that point it barely improves. What is left to build is the thing you talk to. The interface of the next decade is hardware you own.
A memory object
If you mostly talk to it, it does not need a screen. The voice apps we have already glow softly when they listen, and the object on your desk that glows is a lamp. A personal AI should be a thing like that, something you want to keep in front of you.
Peter Kuhar is building one now, an ambient display he calls SidePulse that slots into a MacBook’s card reader and lights up while the agent works.
One I find interesting is Keunwook Kim’s Memory Object, a glowing thing you hold that stores memory in physical material.
Memory is the right word for it. Yours holds your judgment and the way you would answer, built up over years of living with you. The longer it runs, the deeper it goes. It remembers what you believed five years ago and what changed your mind, and it learns your voice well enough to answer as you. You could no more swap it for a fresh one than swap a friend.
A model that spends a whole life beside someone becomes a record of that mind. The person dies and the memory object does not, and you can still talk to it. You could argue with a mind that lived beside Aristotle, or ask a grandmother who died before you were born what she was like at twenty, a living biography you can talk back to.
Only an owned mind lasts that long. A rented one vanishes the day the company sunsets the model or your subscription lapses. An owned one becomes an heirloom, a museum of minds your family and the world can still visit.
Decentralized
Even an owned mind has a ceiling. A model that only ever sees your data can only get so good. Ramesh Raskar’s work on split learning is the way around it. Many people train one model together while each keeps their raw data on their own device.
Your AI learns from your life and passes along the lesson while the data itself never leaves you. Everyone’s AI gets a little smarter from everyone else’s. We rent intelligence and give away our data today, and the owned version hands both back.
The more people who join, the stronger it gets, and no single company sits in the middle with an off switch.
A new internet
Once everyone has their own AI, they start talking to each other.
Your AI messages a friend’s AI. You each say what you need, and the two of them settle the details. Human to agent to agent to human.
Businesses have agents too, and yours talks straight to theirs, with no forms and no website to click through.
For that to work at scale, agents have to find each other, and they have to trust and pay the ones they find. That layer of the internet barely exists yet.
MCP, from Anthropic, already lets an agent plug into any tool or service. Google’s A2A lets agents introduce themselves and hand work to each other, and its AP2 extension lets them pay. Coinbase’s x402 does the same with stablecoins, switching on a corner of the web that was set aside for payments and left unused for thirty years.
The one piece still missing is discovery. MIT’s Project NANDA is a registry beyond DNS for an internet they expect to fill with billions of agents that find and verify each other.
The internet got built this way once before. The standards that moved data between machines and loaded a page in your browser never belonged to any single company, and that is what made the web everyone’s. Tim Berners-Lee saw it coming in 2001, software agents roaming the web and carrying out tasks for people. He was a quarter century early, waiting on models good enough to make it real.
If a few companies own this layer, they own the next internet the way they own the last one. If the agents belong to the people they serve, the new internet belongs to everyone. That is the fork we are standing at.
We have a decade to do something big.