Shanghai has finally cleared up over the past two days, with temperatures approaching 14 degrees. This kind of weather is perfect for sitting by the window, drinking coffee, and writing some code—or… just watching AI write it for you. Coincidentally, while organizing backend materials today, I saw people discussing the pros and cons of different AI programming software. Let’s talk about these two bizarre tools in the tech circle that have recently gone to absolute extremes.
A “Personal Assistant” That Might Wipe Your Hard Drive
You might not know this, but there’s a viral project in the developer community recently called OpenClaw. Its official community has an extremely hardcore warning: strongly advising against installing it on your main computer.
At first glance, you might think it’s some kind of malware, but it’s actually an incredibly powerful AI programming assistant.
It is completely different from the AIs we usually use that just offer code suggestions alongside an interface. With OpenClaw, you don’t even need to open any programming software. You can directly send it a message on WeChat, WhatsApp, or Discord saying: “Go write a web scraper script for the weather and run it.”
You can think of it as a cyber-worker sitting inside your computer. Since it’s doing the work for you, it has the permissions to modify your files and execute various terminal commands. This means that if it misunderstands your instructions, it is highly likely to just nuke your core system files. Therefore, the standard operating procedure for everyone right now is to spin up an independent lightweight application server on cloud platforms like Tencent Cloud, lock it in an absolutely secure remote environment, and control it safely from a distance using a smartphone.
Look at this big red robot icon; it somewhat carries an “I might break things” vibe.
Making AI Not Only Write Code, But Click the Mouse Too
In stark contrast to this wild approach that completely disregards the user interface is another phenomenal tool released by Google late last year—Google Antigravity IDE.
Its foundation is the familiar VS Code, but its biggest selling point is giving the AI massive autonomy. There’s a really interesting detail: it actually has a complete Chrome browser built right into the software. At first, I was puzzled—who doesn’t have a browser on their computer these days? Is it really necessary to cram another one into a coding application?
Later, I realized that this browser isn’t meant for humans at all; it’s prepared for the AI.
After the AI finishes writing a web application for you, it automatically opens this built-in browser, clicks the buttons itself, takes screenshots, and checks if the code it wrote has gone off track. It feels like you didn’t just hire a chef, but you also specifically hired a food taster for them. All you need to do is sit in front of the software’s built-in “Agent Manager” view, sip your tea, and watch several AIs bustling about in different files simultaneously.
The interface looks much more refined; the dense panels are all for monitoring the AI at work.
An 18-Year-Old Retro Easter Egg
Going a bit off-topic here, why did Google name it Antigravity?
Actually, there’s a piece of extremely hardcore geek romance hidden inside. Back in 2008, when Google first launched the Google App Engine, the team had a sudden whim a few weeks before the code freeze and insisted on adding an easter egg. They quietly added a module named antigravity to the Python runtime environment.
If you type a line of import antigravity in Python, your browser will automatically pop up a retro geek comic called xkcd. In the comic, a person is floating in mid-air, and another asks how he is flying. He proudly replies, “Because I used Python!”
Now that Google has directly named its top-tier AI programming tool Antigravity, the subtext couldn’t be more obvious: In the past, writing code in Python was so easy that it made people feel like they could defy gravity; now, with AI, you don’t even need to type the code—just take off on the spot.
Alright, I’m Off to Eat
These two approaches are currently heavily debated in the community. One group is trying every means to make tools incredibly refined and visual, while the other believes that if an issue can be solved by sending a text, they absolutely refuse to touch a keyboard.
As for myself, I only really figured out the tricks behind these two tools while organizing my notes today (facepalm). Actually, thinking about it carefully, if OpenClaw can understand commands in chat apps, could it also help me reply to messages from demanding clients on WeChat? I’m going to look into this terrifying idea, but I’m hungry, so I’m logging off for now.
References:
- How to Run OpenClaw with DigitalOcean
- What programming languages are used to build OpenClaw
- Google Antigravity – Wikipedia
- import antigravity – The History of Python
- 353: Python – explain xkcd
—— Lyra Celest @ Turbulence τ.
