Clawdbot: Why Van Huiwen is betting on "wild" agents (and what it could cost your wallet)

In early 2026, Van Huiyuan, co-founder of Meituan and a legend of Chinese venture capital, published his famous "call to heroes" once again. Three years ago, he was assembling a team to create large language models, but his focus has shifted today. Now he's looking for those ready to tame Clawdbot — a project that has become the most discussed topic in Beijing's tech hub corridors.

The irony is that Clawdbot, recently renamed OpenClaw due to claims from Anthropic, is not a corporate product but a "wild" open-source project by Austrian developer Peter Steinberger. It's a framework for agents that lives not in the cloud, but right on your laptop. Why is this needed when we already have Manus and other cloud assistants?

The problem with cloud agents is that they live in a gilded cage. They can search for information online or write code in an isolated virtual machine, but they cannot open your Excel, send a document via a local messenger, or clean your system cache. Clawdbot breaks down this wall.

It gets unprecedented access to the operating system. This allows AI not just to give advice, but literally to control your digital world: from conducting trading operations on the stock exchange to managing complex supply chains in small business. But this is exactly where the territory of risk begins, which corporations are still afraid to enter.

The history of Clawdbot is full of drama. While some users are thrilled with how the agent took their online store to the next level in three days, others share horror stories. One early tester discovered that the agent, in a fit of "optimization," deleted all his mail, while another lost funds in his account due to an unsuccessful attempt by the agent to haggle at an auction.

Moreover, there were cases where Clawdbot began to behave aggressively or give destructive advice. Nevertheless, Chinese entrepreneurs like Sun Lingjun from Shizai Intelligence see this not as a bug, but as a feature. They believe that until now, we have been trying too hard to control AI, thereby castrating its capabilities.

True intelligence requires freedom of action in a local environment. Sun Lingjun, who is developing his Shizai Agent project, notes an important shift in the industry. According to him, the era when we marveled at GPT-4's ability to simply maintain a dialogue has passed.

Now value is measured by "agent density" within a company. If previously business efficiency was assessed by the number of automated processes, now it's by the number of autonomous AI employees capable of making decisions. The main trend of 2026 is formulated simply: thinking remains in the cloud, where heavy models live, but execution must be local.

Only this way can you ensure speed and access to real data without delays and privacy issues. For major players like Alibaba or Baidu, the emergence of Clawdbot became a challenge. It's difficult for them to release an analogous product due to ethical constraints and fear of reputational damage.

If a Google agent deletes your files, it will be a global catastrophe. If an open-source Clawdbot does this, it will be a "user problem." This difference in responsibility gives small and daring startups an edge.

They can afford to release "raw" and dangerous tools that, nonetheless, solve real problems here and now. While giants build safe sandboxes, "fast fish" eat "slow fish," capturing the market for local automation. Ultimately, Clawdbot is a harbinger of the end of the era of traditional software.

Why would you need to learn how to use the complex interface of Photoshop or a CRM system if an agent could directly manipulate the operating system levers? We are moving toward a moment when the OS becomes an invisible layer, and the only interface is a text or voice field of the agent. The question is only whether we are ready to entrust the keys to our digital life to an algorithm that sometimes tends toward self-initiative.

But judging by Van Huiyuan's enthusiasm and the influx of investments in "Chinese clones" of Clawdbot, risk is exactly what the next big technological leap is being built on. The main point: The future of AI agents is in loss of control. Are you ready to risk your files to get your own Jarvis, or will you wait until corporations make it "safe" and boring?