You're still pretending to be OpenClaw, but this company has already built an AI computer.

Author: David, Deep Tide TechFlow

While you’re still struggling to figure out how to install OpenClaw on your computer or worried about whether it can run fully automatically, one company has already built a machine that comes pre-installed with everything—an AI computer that runs 24/7 without shutting down.

This company is called Perplexity, valued at $20 billion. Yesterday, it held a launch event in San Francisco, quickly sparking buzz in the tech community online. The product name is also quite bold:

Personal Computer.

Buy a Mac mini, install their software, connect to the internet, and you have a never-shutdown AI computer. No need to type commands, set up environments, or find API keys yourself.

Perplexity has integrated 20 AI models—from Claude to Gemini to GPT—each specializing in different tasks, and they handle the work accordingly.

All you need to do is tell it what result you want.

Image source: Perplexity Developer Conference, CEO demonstrating the Personal Computer

Embedding AI Experience into the Operating System

Strictly speaking, Perplexity hasn’t built a computer. What they’ve done is:

Integrate AI directly into a Mac mini’s system.

Buy a Mac mini, install Perplexity’s software, connect to the internet, and this Mac is no longer just an ordinary computer. It runs 24/7, connects with your various work apps, and automatically performs tasks based on your preset rules.

In the demo at the launch, someone gave it a command: filter candidates with SwiftUI experience from the resume database, and simultaneously send an email to investors with a project briefing link.

Two tasks, one command, completed simultaneously. When a customer sends an inquiry email, it drafts a reply in your usual tone; while you’re in a meeting, it updates sales data in the background; while you’re sleeping, it keeps running.

I know your first reaction might be, isn’t this what OpenClaw does? What’s the difference?

In the past two years, there have been two main ways for ordinary people to access AI. One is cloud-based: open a browser, find ChatGPT or Claude, type, wait for the answer, and copy the result for your use. The other is local: install tools like OpenClaw, tinker with environment setup, and let AI operate your computer.

Both paths share a common point—you have to actively seek out AI.

Perplexity aims to do something different: you don’t have to look for AI; AI is already inside your computer.

It directly manipulates your files, emails, calendar, and applications. You don’t need to switch to any “AI interface” to give commands. You don’t need to know which model is running behind the scenes, how tasks are broken down, or how much cloud computing power is used.

You only see that things get done.

This Mac mini doesn’t require a person sitting in front of it. Two weeks ago, Perplexity launched a cloud system called “Perplexity Computer,” with 20 AI models on standby—Claude handles reasoning, Gemini handles research, GPT manages long texts, each with its own role.

Now, the newly released Personal Computer embeds this entire capability into your desktop Mac, turning it into a self-operating computer.

Wrapping is Justice

At the same time, the company’s CEO Aravind Srinivas said something at the launch that I think perfectly captures the product’s essence:

“Traditional operating systems receive commands; AI operating systems receive goals.”

This statement explains why this news has dominated half the tech scene yesterday.

It’s not because another AI product has emerged—this week alone, ten new AI products have launched, and everyone is numb to that. It’s because it shifts a word and offers a more compelling narrative:

Personal Computer.

This term has remained unchanged in meaning since IBM defined it in 1981—45 years ago. You buy a machine, install an OS, and run software yourself. Now Perplexity says, a personal computer shouldn’t be a machine you operate; it should be a machine that does work for you. You’re not the user; you’re the boss.

This narrative hits the hottest track in 2026: AI Agents. OpenClaw has ignited the open-source community, and everyone is betting on the same thing—that AI must shift from “answering questions” to “completing tasks.”

Perplexity also has the narrative capital.

Founded in 2022, its founder Aravind Srinivas previously worked at OpenAI, Google Brain, and DeepMind. The company’s initial focus was simple:

Ask a question, and it uses AI to search, synthesize answers, and cite sources. Think of it as an AI-powered Google, but instead of ten blue links, it gives you a direct answer.

This product hit the right timing. In less than two years, its valuation soared from $500 million to $20 billion, raising over $1.5 billion in total funding, with investors including Nvidia, Bezos, and SoftBank. Its annualized revenue from late 2024 has grown from $80 million to around $200 million.

But Perplexity has a notable—and controversial—feature: it doesn’t develop large models itself.

It orchestrates models from others. Claude is from Anthropic, Gemini from Google, GPT from OpenAI. Perplexity acts as an intermediary layer—arranging these models, wrapping them in its own interface, and selling to users.

In the industry, such companies are called “shell companies.”

But looking ahead to 2026, the term takes on a different flavor. The biggest AI acquisition this year was Meta’s purchase of Manus for billions, which also uses other companies’ models. OpenClaw on GitHub has 140,000 stars, and its underlying models are still Claude or GPT APIs.

Almost no one trains their own models in the AI Agent race. Everyone is “shelling.” The difference lies in how well the shell is made and how many are willing to pay.

Perplexity’s shell now costs $200 a month, its most expensive Max subscription tier.

In February this year, it eliminated its advertising business and shifted entirely to a subscription model. Executives said ads could undermine user trust in search results. As a shell company, it relies entirely on product experience to monetize, not advertising revenue.

Embedding this experience into Mac mini is just the first step; it will also expand to more platforms later.

Contractor’s Dilemma

Shelling can be justified—provided the underlying model providers don’t start doing the same thing themselves.

Anthropic released Cowork, Google is pushing Gemini Agents, and OpenAI’s Operator is heading in the same direction. The models that Perplexity orchestrates are becoming its competitors.

It’s like a contractor whose workers are borrowed from other companies, and now those companies say: we’re also taking on projects.

More complicated are the legal issues of doing the work.

Forbes, The New York Times, and Dow Jones have sued Perplexity, accusing it of scraping copyrighted content. But these are not the most serious issues. Last week, Amazon obtained a federal court temporary restraining order to stop Perplexity’s Comet browser from automatically shopping for users on Amazon. The court found that: Perplexity might have violated the Federal Computer Fraud and Abuse Act.

Why? Because when Comet places orders for users, it didn’t inform Amazon that AI was operating, not a human.

Think about this in the context of the new Personal Computer: a company court-judged to potentially use AI to impersonate human operation on a platform now asks you to open all your local files, emails, and calendars to it, running 24/7.

And there’s another less-discussed number.

Perplexity’s US website traffic increased by less than 4 million visitors from February 2025 to February 2026. During the same period, Claude’s web users quadrupled. Its initial differentiation was AI search, but now ChatGPT, Gemini, and Claude can all do that.

I’m not saying Perplexity can’t succeed.

Shelling out for a $20 billion valuation is itself a skill. But this company is also competing head-on in search engines, browsers, email assistants, cloud agents, and local OS—each line challenging giants…

However, the contractor’s dilemma may not be about competition but about when it will be acquired.