Recently, an AI assistant called OpenClaw has become popular all over the world, and netizens have given it a nickname - "Lobster".

It's not an ordinary chatbot, but an open-source AI tool that can directly operate a computer. With it, AI can automatically complete various complex tasks like your "digital twin": reading news, replying to emails, writing code, and even placing orders and buying and selling stocks online.

In just a few weeks of launch, OpenClaw has become the fastest-growing open source project in GitHub's history. Domestic large model companies also quickly followed suit - Kimi, MiniMax, etc. have announced that they will adapt to "Lobster", so that it can run on its own platform.

From "only chatting" to "doing hands-on work", the boundaries of AI's capabilities have once again made a leap.

The competition for large models has officially entered the second half

At the beginning of March, the downstairs of Tencent's headquarters in Shenzhen was as lively as a market - engineers set up booths in the North Square to help users install the AI assistant "Lobster" (OpenClaw) for free.

There was a long queue at the scene: some people were holding NAS storage devices, some were carrying MacBooks, and some were carrying mini hosts, reminiscent of geeks queuing up to brush Android ten years ago.

In fact, not only Tencent, but also major technology companies are quietly building their own "lobster": Xiaomi is internally testing its own AI agent MiclawAgent, intending to deeply integrate it into the "whole ecology of people, cars and homes" - in the future, your mobile phone, car, TV, and air conditioner may become the "hands and feet" of AI, listening to its command to work.

Earlier, Baidu App already supported one-click invocation of OpenClaw: users can wake up Baidu Intelligent Cloud directly through the search box or message center after deploying it.

Nowadays, cloud vendors have begun to "set up stalls" to promote, and terminal vendors have stuffed AI agents into the operating system. From "being able to chat" to "being able to do things", AI is really starting to enter our devices, lives, and workflows.

"Mission trajectory data" has become the most sought-after resource

For the next generation of large models to truly "do things", it is not enough to just chat - it needs to know how humans complete tasks step by step in the digital world. This data, known in the industry as "Trajectory Data", is now the most sought-after resource.

For example, when you ask AI to "book a ticket to Shanghai tomorrow", it goes through a series of actions: understand your needs, search for flights, open the booking website, fill in the information, select a payment method, and complete the payment. Each step of the operation will leave a record, and these records are stringed together to form a complete "task trajectory".

For AI agents, this data is much more valuable than ordinary text — because it teaches AI real-world logic of action, not just language.

But here's the problem: this data is hidden too deeply! They are scattered in various apps, enterprise systems, and closed software, and even search engines cannot be crawled.

AI agents like OpenClaw and Xiaomi's MiclawAgent, which are deployed on users' computers or mobile phones, can penetrate into these "forbidden areas" and quietly record every interaction like a "data detector".

Alan Feng, community manager of OpenClaw China, put it bluntly: "Users install OpenClaw to make it work automatically; But the real value is that you help it learn how to do it in the process of 'assigning tasks' and 'correcting mistakes'. ”

In other words, you think you are using an AI assistant for free, but in fact you are working as an "AI coach" for a large company for free. Every time you click, modify, or retry, you're generating high-quality training data.

Once this locally generated "trajectory data" is uploaded to the cloud, it will be used to fine-tune and enhance the large model, making it smarter and more capable - just like Tesla relies on millions of electric vehicles to collect real driving data on the road and eventually develop a powerful autonomous driving system.

So, don't underestimate this "lobster" craze.

Big factories are desperately promoting local AI agents, not to give you tools, but to seize the "operation entrance" of the AI era - your mobile phone and computer are becoming their "data collection vehicles".

The entrance war is reincarnated

As users become more and more accustomed to saying just one sentence (such as "book me a plane ticket" or "buy a bottle of soy sauce"), the whole operation method changes: you no longer open Ctrip, Taobao or Meituan yourself, but hand over the task directly to the AI assistant, which decides which platform to use, which service to transfer, and which payment path to take.

This means that the role of the app is being redefined.

They will not disappear, but will change from "protagonists" to "tool people": just nodes that provide services, not portals for users to actively access.

The real entrance becomes the AI agent that does things for you. Therefore, it is still grabbing "whose app is installed on the homepage of the mobile phone", which is outdated.

The real battlefield is who can become the "digital butler" that users trust the most - the underlying agent who directly listens to your instructions and controls the overall situation. Once a giant has its agent firmly stationed in your phone or computer, it has a kind of superpower: "intent distribution rights".

For example: if you say "order a takeaway", it may be pushed to its own platform by default; When you say "book a hotel", it may prioritize completing the transaction with its own payment system. In this new "walled garden" built by AI agents, the once glorious super apps (such as WeChat, Douyin, Taobao) may slowly become "pipes" that only provide background interfaces - users cannot see them, brand value is weakened, and traffic dividends disappear.

This is why all major manufacturers are so nervous and crazy about agents. Everyone wants to be the "commander-in-chief" who stands between the user and the world, not the "waiter" who is commanded.

OpenClaw rehearses AI from dialogue to execution

OpenClaw's sudden popularity may not be just the birth of a popular tool - it is more like a signal of a turning point.

The real change is that AI is evolving from a "chatting assistant" to a "working system". In the past two years, everyone has been fighting for who is smarter and more accurate; Now, more and more companies are beginning to think: How to make AI really do things?

Once AI can perform tasks stably and reliably – such as booking tickets, writing reports, managing schedules, and buying things – the entire structure of the Internet may be reshaped. Users no longer need to open the app one by one, but directly say to the AI: "Help me get this done." ”

The apps and services behind it will be automatically called and retreated to the background to work silently.

In this new world, AI agents are like a new "operating system" that connects you to all digital services. Looking back at the history of technology, every major change often begins with a seemingly inconspicuous starting point: Android was originally just a niche system for geeks to play with; When the WeChat official account was first launched, it was just a simple function for posting articles; When the mini program was first launched, many people thought it was a lightweight web page. But later, they all became the cornerstone of a new platform and a new ecology.

If the future is really the "Agent era", then today's OpenClaw is likely to be one of the first remembered names - just like the Android or mini programs of the past.

For the Chinese Internet, we may be standing on the eve of this storm: an AI revolution from "moving your mouth" to "doing it" has quietly begun.