Alibaba rolls out JVS Claw AI agent for smartphones

Alibaba (BABA) just rolled out a new mobile app called JVS Claw that lets people launch OpenClaw AI agents in minutes. No coding needed.

That move puts Alibaba head to head with Baidu, Tencent, and Minimax in a fight to control how everyday digital tasks get done.

JVS Claw lets anyone spin up AI agents and assign them tasks like online shopping or trip planning. It works much like Baidu’s OpenClaw Android app, which is also built around autonomous agents.

The disruption behind the news: Agentic AI is becoming China’s next platform war.

Alibaba understands how disruptive OpenClaw is, and other AI agents are.

When the interface to access the internet changes, the companies that capture traffic, transactions, and data must change too.

Right now most AI usage still looks like search. You ask questions and read answers.

Agentic AI turns that model into action.

Instead of asking how to book a train ticket, the AI books it. Instead of searching for a product, the AI purchases it. Instead of browsing ten apps, the AI coordinates them.

That’s a major structural shift for the internet economy. Platforms built on clicks and ads could lose leverage if agents start making decisions for users.

E-commerce marketplaces could see AI agents automatically compare vendors and optimize purchases. Travel sites could become background infrastructure if agents handle bookings directly.

China’s tech giants know this. That’s why Alibaba, Baidu, Tencent, and Minimax are racing to lower the barrier to entry.

The strategy is straightforward. Turn AI agents into a consumer habit before the long term business model is fully defined.

Token economics explain some of the urgency. Every AI action consumes tokens, which represent computing work done by the model. More tasks mean more tokens used and more revenue flowing to the platform operating the model.

If even 10% of China’s 1B internet users start running agents daily, token demand would surge. A simple workflow of five tasks per user could create billions of automated requests per day.

But the deeper incentive may be transaction capture, not tokens.

If an agent books a $200 flight or buys a $50 product, the platform controlling that agent can take a 1% to 2% routing or referral fee.

At 100M users running just one $100 transaction per week, that’s roughly $5B to $10B in annual transaction flow before advertising even enters the picture.

A few days ago we showed you that's why the agent layer is the reason OpenAI can finally become profitable.

What to watch next

First watch user behavior.

The “raising lobsters” craze, spreading among students and retirees matters more than the technology itself.

Raising lobsters, or yang longxia, is a trend that involves people experimenting with small AI agent tasks and sharing the results online.

That kind of behavior signals habit formation, which is often the first sign that a new platform is gaining power.

Second watch integration deals.

Agents only become valuable when they connect to services like payments, logistics, and travel platforms.

Third watch compute economics.

If token costs fall fast enough, agents become cheap enough to run constantly.

One subtle constraint could shape how this market gets divided up. It’s compute.

If a typical agent workflow costs even $0.01 in inference, then 5 daily tasks across 100M users adds up fast. That’s about $5M a day in compute spend, or roughly $1.8B a year.

That kind of cost means large-scale agent adoption may only work for companies that already run huge data centers. Or for those willing to subsidize usage because they expect to make money later through payments, commerce, or other downstream transactions.

And the timeline may not be far off.

Within 24 months, millions of Chinese users could have persistent AI agents handling routine digital tasks for them, especially as ecosystems form around platforms.

If that happens, the internet’s main interface stops being a browser. It becomes an AI agent.