Is the AI assistant that has caused the price increase of Mac Mini really that powerful?

Recently, social media has been flooded with posts about a project called Clawdbot. The hype is so intense that it feels like I've been transported back to the afternoon when Gemini 3 Pro was just released.

It's an open - source project on GitHub that has already amassed over thirty thousand stars, and the number is still growing.

In simple terms, its purpose is to allow an AI to control your computer to do anything through conversations.

Some people have used it to create a digital employee that monitors the market 24/7 and can place orders autonomously to make money, achieving fully automated trading around the clock.

Some claim that they used it to negotiate with dealers and managed to get a $4200 discount when buying a new car.

A foreign tech blogger is even more obsessed. He said that he consumed 180 million Tokens with Clawdbot in a week. His final conclusion was: After experiencing this superpower, I can never go back.

Wow, isn't this just like Jarvis in Iron Man? I can just sit on the chair and give commands, and the computer will do all the work for me. Could it be that the future has really arrived...

Since it runs well on Mac mini with low power consumption and small size, the price of Mac mini has even gone up.

So, I dug out the old Mac in our editorial department and spent an afternoon installing this thing.

After using it for an afternoon, both I and the computer were on the verge of overheating. Let me give a conclusion first: It is indeed useful, but some people may have over - hyped it.

Before talking about my experience, let me give a brief introduction to why this thing has become so popular for those who are not familiar with it.

To put it simply, Clawdbot is different from chatbots like ChatGPT. It doesn't just talk; it's a real - working Agent.

It can operate your computer, read and write files, send emails, control the browser, and even open VS Code to write code for you and run tests on its own after writing.

Moreover, it supports multi - channel access. That means if you send a message in communication apps like Telegram or WhatsApp on your phone, it can start working on your computer without you being next to it.

It also claims to have permanent and unlimited memory. If you asked it to do something a few days ago, it will still take the initiative to do it for you a few days later. From the description, it really seems like a real human intern. Can it really be willing to be exploited?

It sounds great. Let's test it out. I assigned it three tasks.

First is the classic desktop cleaning task. These days, various AI tools have emerged, like Cowork and Openwork. My folders are in a complete mess.

So, I sent it a message on Telegram: "Help me sort the screen - recording folder by file type."

Then Clawdbot started working. It first scanned the folder, created several categories such as screen recordings, document materials, screenshot images, and others. Then it started moving the files one by one.

As a result, it worked really well. It sorted everything neatly, and the files were correctly categorized.

It took about a minute. To be honest, I could probably do it manually in about a minute too, but the advantage is that I don't have to think.

In addition, Clawdbot can also achieve the "AI commanding AI" nesting operation, which means it can command other AI tools to work for you.

For example, I can ask it to go to Claude Code and write a Python script for me to batch convert video files into GIFs.

The process went smoothly. The code worked without any bugs, and it was really convenient. It used to be much more troublesome to convert a video into a GIF before.

The third task was to help me post on Xiaohongshu. This is the task I was most eager to try because it involves browser operations and social media posting. If it succeeds, we can achieve fully automated social media management in the future.

I gave it a simple request: help me post a message on Xiaohongshu with any content.

Then... it got stuck.

The reason was that I hadn't logged in to the web version of Xiaohongshu.

Logically, when it got stuck, it should have asked me for my account and password, but it didn't. It just waited there.

I asked it what it was waiting for, and it didn't reply. Then I thought of the scariest possibility.

I checked my account, and it had used up all my Tokens.

I just recharged $5, and it used it up in half a task?

Later, I had to reluctantly recharge another $5 and switched to the cheaper GLM model... Now I don't have to worry about my wallet, but the performance seems to have declined...

I logged in for it and opened the page for it. Then I had to keep urging it like whipping a top to make it move forward.

After I urged it repeatedly, it finally completed the task. However, the picture was really ugly, with a huge gap...

After completing the three tasks, I have a clearer understanding of Clawdbot: it is indeed useful, but its performance depends largely on the underlying model. For example, when I switched to GLM, the effect was slightly worse than that of Claude Opus.

Moreover, if you want to do some complex tasks, your wallet may not be able to bear it. It may cost dozens or even hundreds of dollars a day. It's not like hiring an intern; it's more like taking care of an ancestor...

If you just want to have some fun, the installation and configuration threshold of this thing is far from being as easy as many media claim. You need to at least understand some English and code because the error messages are all in pure English, and many of them are quite confusing.

I spent a long time dealing with it to get it working. If you're a complete newbie, you'll probably be discouraged during the configuration process.

Even if you can get it working, for us domestic users, it can only operate browsers and desktop applications. Apps like WeChat, Douyin, and Taobao, which we use every day, have strong risk - control mechanisms. If you use AI to operate them, you may get a warning at best or have your account blocked at worst.

Finally, there are the inevitable security risks. It can work because you give full control of your computer to an AI, which means you're handing over your account passwords, bank card information, chat records, confidential files, etc. to the AI.

For example, a user asked Clawdbot to help buy a car. It automatically opened the browser to fill out the form and entered the real phone number. Then that guy started receiving a lot of harassing calls.

If someone hides a malicious command in a web page and Clawdbot reads it and helps transfer money, send emails, or delete files, it won't be that simple. So, if you want to try it, either use an old computer you don't use often or open a virtual machine. Don't use it on your main computer without protection.

Of course, I also have some subjective feelings: for me personally, I can finish many tasks in 30 seconds by myself. If I let the AI do it, I have to describe the requirements, wait for it to execute, and check the results... The process becomes much longer.

Moreover, there's a chance that it may fail. For example, Cowork has caused some problems recently