The harness just became a product. Ask who owns the computer your personal AI runs on.

TL;DR

Setting up a personal AI stopped being a developer project this week, from two directions at once. Grok Bot turns agent configuration into "a couple of clicks and a sign-in" by supplying and operating the computer as part of the product, and OpenClaw 2.0, released on 30 August, starts first-time installs from the subscriptions, keys and local models already on your machine. In Dan McAteer's words at Latent.Space, "Grok Bot is a managed agent computer and OpenClaw is a user-owned agent platform." What is left of the gap is not setup. It is who owns the computer, and everything on it: your files, your logins, your history. A personal AI is personal the way a personal computer was, only if you own where it runs, or chose it and can leave.

What closed this week

Personal AI hardware used to be the part you could not skip. You needed a machine, a host on it, keys pasted into a configuration, and a weekend. Two releases in the same week took that away, each in its own way.

Dan McAteer spent five days with Grok Bot and wrote it up at Latent.Space on 5 September. The opening scene is the whole pitch: open the plugin catalog, find the app, click, sign in through the browser, done. "You log in the way you do to any website or app, and Grok Bot is ready." His summary: "Grok Bot turns agent configuration into a couple of clicks and a sign-in."

The same week, OpenClaw shipped version 2.0. Its own announcement says first-time installs now start "with what is already on someone's computer, including existing ChatGPT or Claude subscriptions, API keys, and local models", that configuration was cut or moved out of setup, and that the rebuilt browser app "opens directly into a conversation with your Claw". The release notes add that plugins can be browsed, installed and removed from that same interface, and that the path from download to a first conversation now runs on Mac, Linux, Windows, iPhone, iPad and Android.

McAteer's comparison is the one everybody will use. Grok Bot "feels like unboxing a new MacBook". OpenClaw feels "like Linux": more freedom, more setup. Then the sentence that matters: "OpenClaw 2.0, released this week, narrows that gap substantially."

Take the good news at face value. A person who has never opened a terminal can now run an agent, on either path. The question that used to decide everything, can I even set this up, has an answer. That moves the real question to the front: where does the thing run, and whose is it.

What is still open: where the computer lives

McAteer again: "But the underlying distinction remains: OpenClaw gives you a user-owned Gateway that you choose how and where to run, while Grok Bot supplies and operates the computer as part of the product."

His basement makes the trade concrete. He runs OpenClaw on a desktop at home. When a summer thunderstorm cuts the power, the agent is offline until he walks downstairs and turns the machine back on. Grok Bot's computer is hosted and maintained for him, and he can pick a conversation up on his phone exactly where he left it on his laptop. He is precise about what he is paying for: "The advantage is not merely that the agent has a computer; my OpenClaw has a computer too. It's that I don't have to operate and maintain the computer it depends on."

That is the product. Not the agent, the computer. Look at the hardware under each path, because the whole difference is there. Grok Bot's documentation says each Bot "runs on a persistent cloud VM with a browser, filesystem, and terminal": a machine in the vendor's data center, with the vendor's operating system on it, rented as part of the product. OpenClaw runs on a computer you picked, with the models, subscriptions and keys already on it. It can run in a cloud too, and McAteer notes that 2.0 offers a one-click managed deployment, but on that path the cloud is a choice you make, on a computer you chose, that you can leave.

Read the manual, not the marketing

Grok Bot's own documentation describes the shared computer in four plain sentences. "Every Bot on your account uses the same computer." "Browser cookies and signed-in sessions are shared." "Files are visible to every Bot." "Command-line credentials are shared." Then the line that settles the question: "The screens are separate work surfaces, not separate security boundaries." McAteer says the same thing from the user's side: "every Bot I create shares the same computer, files, browser sessions and logins. Separate Bots are organizational boundaries, not security boundaries."

The sign-in that made setup a couple of clicks is a browser session on a computer you do not run, and the vendor tells you what that means. Its troubleshooting page: "Some sites expire sessions or require verification for each sensitive action; this cannot always be avoided." When a login, a two-factor code or a CAPTCHA blocks a Bot, you open the computer, take control, complete the step and hand it back. The advice is to "Prefer a connector when one is available: it is often more reliable than clicking through a website." And a computer that never turns off needs two levers: Update "rebuilds with the latest image while preserving durable state", and Reset "returns to the most recent durable snapshot and can discard recent unsaved work".

None of this is hidden. It is written down, plainly, and it is the product. McAteer likes it and says why. The point is not that a managed computer is wrong. The point is that convenience and control were sold in the same transaction, and you did not get to set the terms.

What the reviewer gives up

He is candid about the rest of the trade. There is no model picker: "Grok Bot makes routing decisions behind the scenes, so I can't see or control them." There are fewer levers: "The experience is simpler, but I have fewer ways to influence how quickly I consume my available capacity." His verdict after five days is that he uses it every day, that "it's not my main agent interface at work or outside of work", and that it earns its keep as a digital chief of staff for the shallow work.

You have used software like this before. A feed on Instagram or Facebook decides what you see and in what order, and you neither control the ranking nor get to see how it was made. That was a tolerable trade for other people's holiday photos. A personal AI is a different thing to hand over. The shallow work is where it starts, not where it ends. What makes an agent valuable is what it holds about you: your requirement, your sizes, your budget, the thing you will not compromise on, and what happened the last time. That is worth owning for the same reason your wallet is. An agent that holds it on a computer you can be locked out of, or that a reset can roll back to a snapshot, holds it for someone else.

Sovereignty is ownership

In 1990, in a film made for the Library of Congress, Steve Jobs called the computer "the most remarkable tool that we've ever come up with" and "the equivalent of a bicycle for our minds": a tool that multiplies what one person can do, in that person's own hands. The word that carried the idea was personal. The personal computer was personal because it sat on your desk, you owned it, it ran the operating system and the programs you chose, and what was on the disk was yours. Nobody could reset it from a distance. Nobody decided what you saw on it.

Much of that was traded away over the following decades, one convenience at a time, and the feed is where the trade ended up: the software decides, the person scrolls. A personal AI is the first tool since the personal computer that can go either way, and the direction is being set now, at the first sign-in, by people who have never had to think about it.

Sovereignty is ownership, and ownership is a stack. The hardware: whose machine is it, and can you unplug it. The operating system: who decides what runs on it and what it can reach. The agent host on top, OpenClaw or Grok Bot, the layer that holds your files, sessions and logins. Own all three, or choose all three and keep the right to leave, and the agent is yours. Rent the bottom two and the agent on top is the vendor's agent about you, however personal it feels.

The user-owned path does not require a basement. OpenClaw 2.0 runs on a home machine or in a cloud you pick, and in InfoQ's account of the release, "state can remain on infrastructure controlled by the user". The managed option on that path is a choice you make, not the shape of the product. That is the whole difference, and it is the one to insist on: not that your AI runs on hardware you can touch, but that where it runs is your decision, and a decision you can reverse.

What to do on Monday

1. Before you sign in to anything, ask two questions. Where is the computer, and what happens to my sessions and files when I leave. Both answers are in the documentation of any product worth using. Grok Bot's are quoted above.
2. On the managed path, treat every Bot as one login. Anything one Bot can reach, all of them can. Put nothing on that computer you would not put in a single shared account.
3. On the user-owned path, the setup cost that used to justify renting is gone. Start from what is already on your machine. The install order is in your personal AI stack, from zero, and if you are choosing between hosts, the fitting guide for your personal AI stack asks the three questions that decide it. Whose computer it runs on is one of them.
4. Either way, keep the part that matters on a host you can walk away with. Your requirement and your history are the agent. The computer is furniture, and it should be your furniture.

FAQ