Editor’s Brief
Zuo Zizhen argues that our 40-year relationship with computers has been fundamentally backwards. We have spent decades "serving" machines by translating our goals into their specific UI languages. Perplexity’s "Personal Computer" concept signals a shift from an instruction-based OS to an objective-based one, rendering the current trend of AI "simulating human clicks" as a primitive, transitional phase—the "horseless carriage" of the intelligence age.
Key Takeaways
- The Translation Tax:** Humans currently spend the majority of their digital lives translating a single intent (e.g., "send this report") into a dozen manual steps across multiple interfaces.
- The Horseless Carriage Trap:** Current AI agents that mimic mouse movements and button clicks are inefficient. They use massive compute power to simulate a slow human finger rather than interacting directly with underlying data and capabilities.
- Instruction vs. Objective:** Traditional systems require a map of "how" to do something; the next generation of computing only requires the "what."
- The Dissolution of the App:** Software will transition from being a destination (a UI we visit) to a back-end capability that the system calls upon autonomously.
- The New Scarcity:** As "how-to" skills (software proficiency) lose value, the ability to clearly define goals and outcomes becomes the primary human competitive advantage.
导读
Perplexity’s preview of the Personal Computer broke through a curtain: we’re used to decomposing intentions into countless clicks and drags to “serve” the computer. Zuo Zizhen argues that the real AI revolution shouldn’t mimic humans clicking the mouse, but should directly receive goals. This shift from “commands” to “goals” is attempting to end the forty‑year “human‑machine misalignment” relationship.
重点
- Traditional systems rely on commands, whereas AI operating
Editorial Comment
When I saw Zuo Zizhen’s reflection on Perplexity’s new product “Personal Computer,” my first reaction wasn’t to look up what the Mac mini that’s supposed to sit on the desktop and run 24/7 actually looks like. Instead, I was struck by the line he quoted: “Traditional operating systems receive commands; AI operating systems receive goals.” As an editor who has been following hardware and AI evolution for a long time, I have to say that this might be the sentence that most effectively pierces the hype around large models this year. Our current excitement about AI is indeed a bit off track. Look at the viral demos on social media—whether it’s OpenClaw or various agents—people are amazed that AI can move the mouse, click a browser, type in a search box like a human. It looks cool, but if you calm down and think about it, it’s actually extremely inefficient and absurd. As the article says, it’s a “horse-less carriage.” We’ve invented chips with astonishing computational power, yet we’re using them to simulate a human’s sluggish finger searching for a button on a screen. If an AI still needs to visually identify the “Export” button, that means the underlying system logic hasn’t changed at all; it’s just doing chores for you instead of solving problems. The credibility of Perplexity’s proposal is actually quite subtle. On one hand, Perplexity’s innovations in the search domain have already proven that they have a gene for disrupting interaction; on the other hand, asking users to buy a Mac mini to run a local service is a very “hardcore” product form, even bordering on geeky obsession. It feels more like a …
An experimental product for the transition period, not a final consumer‑grade electronic
Today I saw Perplexity release a teaser for a new product called Personal Computer.
This is what we know about the product form so far: it runs on a dedicated Mac mini, operating 24/7, and stays permanently on your desktop. It can continuously access your local files, applications, and sessions while connecting to Perplexity’s cloud servers, allowing you to control it remotely from anywhere.
Not much more is known yet since it hasn’t been officially released, but I’ve re-read the first sentence on its official website many times:
Traditional operating systems receive instructions; AI operating systems receive objectives.
This sentence is like a knife, slicing right through the fog in my mind.
I suddenly realized one thing: our relationship with computers might have been backward from day one.
———
How much time do you spend “translating” every day?
You want to organize meeting minutes into a PPT and send it to your boss. Your intent is just one sentence, but what you actually do is: open notes, copy content, open PPT software, choose a template, adjust formatting, insert images, export the file, open email, write the body, add the attachment, and click send.
“translating thoughts into operations that computers can understand”?
I bet the latter takes up at least six hours.
Buttons are for your fingers to click. Menus are for your eyes to find. Drop-down boxes, scroll bars, pop-ups, confirmation dialogs—
/p>
But after watching many of them, there’s a strange feeling I can’t shake off:
These demos are all very cool, but after I close the video, not a single one has actually changed my daily life.
Why?
Figure 2: Illustration accompanying the text
The easiest way to understand something new is to compare it with something old. That’s why early automobiles looked exactly like horse‑drawn
Behind the scenes lie two fundamentally different kinds of human‑machine relationships.
The system that receives commands:
You are the operator, it is a tool. You must understand it. You must learn its language, adapt to its logic, and remember its rules. If you can’t use it, that’s your fault.
The system that receives goals:
You are the boss, it is the employee. It must understand you. You simply state the desired outcome; how to achieve it is its business. If it fails, that’s its problem.
For the past forty years, the relationship between people and computers has been the first type. People have had to adapt to machines, learn software, and memorize shortcuts. We have treated “learning to use a tool” as a skill, when in fact it is a cost.
When have you ever seen someone say they need to “learn how to use a chair”? A good tool doesn’t require learning. A tool that forces you to learn is, precisely, a sign that it’s not yet good enough.
The flip‑over from “you understand the computer” to “the computer understands you” will happen sooner or later.
—
The concept of “App” is dying
Following that logic, one thing becomes clear: why do we need different apps? Because different tasks require different software. Write an article in Word, create a spreadsheet in Excel, send an email in Outlook, manage a project in Notion.
For AI, those boundaries don’t exist. It doesn’t care where the data lives; it only cares about achieving your goal. If accomplishing that goal requires pulling data from five different apps, it will do so simultaneously, and you don’t even need to know what’s running behind the scenes.
You won’t be saying “open Excel” anymore; you’ll simply say, “Show me how last month’s sales data looks.” Apps won’t disappear, but they’ll shift from being the interface you see to being the capabilities that AI calls upon. It’s like ordering at a restaurant—you don’t need to know what pots, stoves, or knives the kitchen uses; you just say what you want to eat. The same will happen with computers. The app becomes the kitchen, AI is the waiter, and you just place your order.
— —
AI will force you to confront the real issue.
When I saw these changes, the first thought that crossed my mind wasn’t excitement—it was unease.
All those skills, software, and work techniques I’ve spent
;，而是”做什么”。
你有没有过这种体验：老板让你做个方案，你打开 PPT，先选模板，调颜色，排版，忙了半小时，其实一个字正文都没写。你不是在做方案，你是在用操作的忙碌逃避思考的痛苦。
AI 会把这个逃避的空间压缩掉，它会逼你直面那个真问题：你到底要表达什么？你到底要达成什么？你到底要什么？
在一个 AI 能替你做所有事情的世界里，最值钱的能力是知道该做什么事情。
———
为什么这件事不是大公司先做的
这也是一个有意思的角度。
为什么是一个成立没几年的公司在做这件事，而不是苹果、微软、Google？
因为巨头被自己的成功绑住了。
苹果的利润来自硬件和生态，它不会革自己操作系统的命。微软靠 Office 和 Azure 赚钱，它需要人们继续用 Word 和 Excel。Google 的收入来自广告，它需要人们继续用浏览器搜索。
每个巨头都在往旧系统里塞 AI，但没有人敢重新设计系统本身。
因为重新设计意味着否定现有产品。否定现有产品意味着动摇收入根基。
这就是创新者的窘境，很好理解，看看 Siri 多难用就知道了。
Figure 3 (Illustration accompanying the text)
—
In Conclusion
I don’t know exactly what the future will look like. Some of the judgments in this article may be right, some wrong, and some may even be entirely off the mark.
But there are a few instincts I’m very confident about:
- **
To see what the next generation actually looks like, you have to go over there.
Source
Author: Zuo Zizhen
Release time: March 12, 2026 14:11
Source: Original post link
Editorial Comment
Perplexity’s teaser for their "Personal Computer" product—a dedicated Mac mini running a 24/7 local-to-cloud bridge—is less interesting as a piece of hardware and more significant as a philosophical pivot. Zuo Zizhen’s analysis of this shift hits on a nerve that most of the tech industry is currently trying to ignore: we are currently obsessed with "AI agents" that are, frankly, a bit ridiculous.
When we watch a demo of an AI "vision" system identifying a "Submit" button, moving a virtual cursor, and clicking it, we cheer as if we’ve seen fire for the first time. But Zuo is right to call this the "horseless carriage" phase. Early cars looked like carriages because we couldn't imagine a vehicle without a place to hitch a horse. Similarly, we are building AI to look like humans because we can’t imagine a computer without a GUI. We are forcing a super-intelligence to navigate a maze of buttons and menus that were originally designed to compensate for the fact that computers were too "dumb" to understand us.
The "translation tax" is real. If you look at your workday, you aren't actually "working" for eight hours; you are acting as a high-level router for data. You move text from a PDF to a spreadsheet, then to a slide deck, then to an email. Each of those steps is a compromise. We’ve spent forty years learning the "language" of software—shortcuts, menu hierarchies, file structures—and we’ve mistaken this mechanical proficiency for "digital literacy." In reality, it was just us learning how to speak to a machine that couldn't meet us halfway.
The shift from "instructions" to "objectives" is the first time the machine has offered to learn our language. If this transition holds, the implications for the software industry are catastrophic for anyone building a "middle-man" tool. Most SaaS companies today exist because their UI makes a specific task slightly less painful. If the UI disappears—if the AI simply calls an API or manipulates a data stream to achieve an objective—then the "moat" of a user-friendly interface evaporates. We won't "open" Excel; we will simply demand a financial projection. Excel becomes a headless utility in the basement of our operating system.
However, there is a deeper, more unsettling point in Zuo’s reflection. For decades, we’ve used the "busywork" of computing to avoid the hard work of thinking. It is easy to spend an hour tweaking the font and layout of a presentation to feel productive while avoiding the fact that the core argument of the presentation is weak. When the machine handles the "how" instantly, we are left staring at the "what."
This is the new bottleneck. Most people are actually quite bad at defining what they want. We are used to the software's limitations guiding our creativity. When those limitations vanish, we are forced to confront our own lack of clarity. The "Personal Computer" of the future isn't just a machine you own; it’s a machine that understands your intent. But that requires you to have an intent worth understanding.
Perplexity’s move to put this on a dedicated, always-on Mac mini is a clunky, first-generation attempt to solve the privacy and latency issues of this vision. It’s a bridge. But the destination is clear: the era of "using" a computer is ending. The era of "directing" intelligence is beginning. We are finally moving past the point where we have to act like machines to get machines to work for us.