2026: The Year AI Agents Step Out of the Lab and Onto Your Desktop

From July 17 to 20, 2026, the World AI Conference (WAIC) will take place at the Shanghai World Expo Center. This year's edition sets new records in exhibition scale, debut products, and forum sessions — and "AI Agent" will undoubtedly be one of the conference's defining themes.

Why 2026?

Zhou Hongyi, founder of 360 Group, offered a timeline at the Global Digital Economy Conference in early July: "AI will reshape the world within 5 to 10 years." He also declared 2026 the official "Year of the Agent," predicting that the internet's infrastructure will be fundamentally restructured by billions of "silicon-based agents."

This is not empty rhetoric. The past six months have seen an extraordinary density of developments in the AI Agent space:

• Manus: Launched early this year, this general-purpose AI agent was the first to ignite market imagination around AI that can actually do things.
• OpenClaw: This open-source AI agent project continues to gain traction, with cities like Shenzhen and Wuxi rolling out policies to support its deployment in manufacturing and government services.
• CoPaw: Alibaba's open-source desktop agent tool, released in late February, supports one-click local and cloud deployment, integrating with DingTalk, Feishu, QQ, Discord, and other messaging platforms.
• Cloud providers join in: Tencent Cloud, Alibaba Cloud, and Baidu AI Cloud have all integrated AI Agent capabilities, lowering barriers through simplified deployment workflows.

From Chatting to Doing

Zheng Jichang, dean of Tsinghua University's China Innovation Service Research Institute, offers a fitting metaphor: AI agents are "digital employees that can both think and execute."

What sets AI agents apart from traditional conversational AI is their "closed-loop execution capability." Tell one to "compile this week's work into a report and send it to the department's Feishu group," and it won't just give you text — it will read relevant files, generate the document, open Feishu, and send it. From understanding the task to completing it, no human intervention is needed.

But Zheng is also sober about the current state of the technology. Today's AI agents are "more like junior interns who need careful supervision." While they can handle specific scenarios efficiently, misunderstandings and operational errors remain common. The gap between "usable" and "reliable" has yet to be closed.

On the Eve of a Trillion-Yuan Market

Analysts project China's AI agent market will exceed 3.3 trillion yuan by 2028. That number reflects a fundamental shift: AI moving from an assistive tool to a core productivity engine.

Data from China's National Development and Reform Commission shows that annual shipments of AI-enabled devices — phones, PCs, and other smart terminals — surpassed 100 million units in 2025. AI penetration in key industries has topped 80%. When both devices and industries are "AI-ready," the mass deployment of agents is just a matter of time.

Risks and Challenges

As AI agents enter the workplace, security risks follow. Security experts at 360 have warned about "prompt injection" attacks, where hidden instructions embedded in web content can trick agents into leaking sensitive information like API keys. When agents gain access to email systems, code repositories, and databases, security governance can no longer be an afterthought.

Zheng Jichang advises enterprises to "prioritize building an AI-ready data foundation and governance framework" — a challenge that spans both technology and management.

Ahead of WAIC

In one week, the global AI community will gather in Shanghai. From foundation models to intelligent agents, from world models to embodied intelligence, one theme runs through it all: AI is moving from "thinking" to "doing," from technical demos to real value delivery.

2026 may not be the year AI replaces humans — but it is likely the year we start getting used to collaborating with AI agents. That "digital colleague" on your desktop is already on its way.