Gemini 3 clones Mac Activity Monitor in one afternoon

Gemini 3 clones Mac Activity Monitor in one afternoon

Can an AI write a clone of the Mac Activity Monitor that also runs on Windows and Linux in a single afternoon? According to the story that sparked this discussion, yes: a system administrator tasked Gemini 3 with creating a control panel that condenses into one window what the operating system splits between the process monitor and the system information app. The result, according to its author, works on all three platforms, is written in Go using the Fyne library, and does not reuse data generated by Apple's native tools: it collects it from scratch and sends it via email.

The detail is not minor. The program doesn't just show CPU load, memory, or network estimulante ilegal: it also identifies the device with model, serial number, OS, and IP address, and lists the disks. All of this, according to the initial premise, with fewer windows and fewer clicks than the factory-default tool. The confessed motivation was twofold: solving a technical problem and showing pogre to the boss.

What exactly does the panel generated by Gemini 3 do?

The summary provided by the AI itself describes the program as a car dashboard applied to the computer. Speedometer-style gauges for the CPU, memory graphs, and an estimate of whether the connection is fast or slow. On top of that, a quick spec sheet with the data tech support usually asks for when handling an incident.

The interesting part is where those data come from. The author insists that the program doesn't read what the Activity Monitor or System Information app already outputs, but rather gathers the information independently. That’s what turns the exercise into something more than a pretty interface: it involves digging into the guts of the operating system across three different platforms.

The recurring pattern: AI accelerates, humans correct

The conversation quickly turned to the elephant in the room. Several participants agree that the outcome depends entirely on who asks: you need to know programming and how to explain what you want, because the tool derails easily. The author himself admits it without mincing words: sometimes it goes off the rails, and then you have to compile and debug.

The pattern repeats in the cases shared. A personal bonus calculator that required several prompts and tweaks until it was decent. A basic Python network scanner solved in ten minutes. A list of fifty PlayStation games without action that kept sneaking in action titles again and again, even after acknowledging the error. The shared conclusion is uncomfortable: the more generic the task, the better the result; the more specific or longer it is, the more it gets confused.

Antiestéticar of unemployment and the response of those jumping on board

The thread has its apocalyptic vein. Some warn that in a few years we’ll be queuing for unemployment benefits, while others respond that the queue no longer makes sense: the destination would be different. The comparison with the movie where surplus humans become food, and the image of human batteries, appears without irony.

In contrast, technical profiles respond with pragmatism. The program’s author defines himself as a kind of IT plumber: today he generates code, tomorrow he climbs onto a roof to lay cable or opens a computer to repair it. For him, AI is a tool, not a threat. Others point out that what loses value isn’t typing code, but knowing how to design products, and there the machine remains a disaster.

The bubble, energy, and the leap ahead

Not everything is enthusiasm. Part of the analysis argues that the energy requirements of exponential growth are unsustainable, and that’s the root of the bubble. The most optimistic response doesn’t deny overvaluation: they take it for granted and consider it irrelevant, because the bet isn’t on what AI does today, but on what it might achieve. A machine’s ability to process data puts human capability to shame, and that, they argue, will accelerate entire fields.

The counterpoint comes from the craft side. It’s warned that convenience has a price: programmers who never solve a problem without help lose their instinct, and when AI can’t find a way out, it sticks to the error and drags the user along. The repeated recipe is to break the problem into pieces and know how to do some of them manually.

---

In the end, the anecdote that best summarizes the state of affairs is the sum: 7 plus 7 plus 7 plus 13, and the machine fails by ten. With that margin of error, trusting it with your car’s dashboard is amusing. Trusting it with the engine, for now, is not.