WAIC 2026: AI Agents Take Center Stage as AI Finally Gets to Work

No One Cares About Parameters Anymore

On July 20, the four-day 2026 World Artificial Intelligence Conference (WAIC) wrapped up in Shanghai. The numbers were impressive as always — over 100,000 square meters of exhibition space, 1,100+ exhibitors, 3,000+ exhibits, 351 product debuts.

But something was different this year, and you could see it in how the booths were designed.

In previous years, AI companies put parameter counts and benchmark rankings at the center of their displays. This year, real business use cases took the spotlight. Robots appeared in warehouse logistics and retail stores. AI agents handled end-to-end workflows — booking tickets, processing payments, managing office tasks. When company representatives spoke, they talked about deployment numbers, performance metrics, and customer feedback — not about how many billion parameters their latest model had.

The signal from WAIC 2026 was unmistakable: the AI industry is pivoting from a technology race to an application race. "Getting things done" is replacing "being clever at conversation" as the new measure of AI value.

What Exactly Is an AI Agent?

One word echoed through every hall at WAIC 2026: "agent."

A tally by journalists found that over 10% of the conference's 100+ forums and events were directly about AI agents. Among the ten "Crown Jewels of the Exhibition" announced by the organizers, four were agent products — Baidu's general-purpose agent "Baidu Dazi," China Southern Power Grid's grid-planning agent "Dawate Yunrui," Stepfun's native agent smartphone STEPX Neo, and the conversational protein-design agent MatwingsVenus (Xiaowu).

So what exactly is an AI agent, and how does it differ from the chatbots we've grown used to?

In May, three Chinese government agencies — the Cyberspace Administration, the NDRC, and the MIIT — jointly issued the "Implementation Opinions on Standardizing AI Agent Applications and Promoting Innovative Development," which offered an authoritative definition: "An AI agent is an intelligent system capable of autonomous perception, memory, decision-making, interaction, and execution. It represents a key form of AI products and services."

In plain terms: a chatbot can only "think" and "talk." An agent can also "act." It can autonomously complete a sequence of operations around a goal — searching for information, making decisions, calling tools, and delivering results.

According to AI research institution "Intelligent Hyperparameters," China saw at least 2,770 application-oriented LLM bidding projects in the first half of 2026. Of these, 673 — nearly 30% — were agent-related, with roughly 180 involving agent development platforms.

Luo Chao, head of the institute, put it bluntly: "Market demand is no longer satisfied with simple text generation or conversation. The expectation now is that AI will truly embed into real workflows and complete complex tasks."

The Numbers Don't Lie

If I had to pick just one number to capture the explosive potential of the agent industry, it would be this:

At the 2026 China Internet Conference, Chinese Academy of Engineering academician Wu Hequan stated publicly that by 2030, conversational token traffic in China will drop from roughly 50% of total AI traffic in 2025 to just 12%, while agent and agent-service traffic will surge to 75%.

From 50% to 12%, and from near-zero to 75% — this is not incremental change. This is a structural flip.

IDC China President Huo Jinjie's forecast is even more striking: by 2029, there could be over one billion AI agents deployed globally. For context, all mobile app stores combined currently host only a few million applications.

Gartner's July survey provides another perspective: 17% of organizations have already deployed AI agents, 42% plan to do so within 12 months, and another 22% within the following year. Gartner Senior Research Director Zhang Tong added that the 17% figure likely understates reality due to sampling constraints — his field observations suggest large numbers of enterprises across sectors have already deployed multiple agents, and the trend is accelerating.

A joint report from Salesforce, MuleSoft, and Deloitte Digital points in the same direction: 89% of enterprise teams globally are already using AI agents, with an average of 12 deployed per organization. That number is expected to grow 67% to 20 within two years, and IT leaders plan to allocate 19% of their budgets to AI agents.

Every data set tells the same story: AI agents are not a future trend. They are the present reality.

The Whole Supply Chain Is Moving

The heat in the agent industry is radiating from upstream to downstream at an accelerating pace.

Standards first. In late June, China released seven national standards under the "AI — Agent Interconnection" series, covering architecture, identity codes, identity management, agent description, discovery, interaction, and tool invocation. The MIIT's March plan showed that at least 12 more industry standards for agents are in development — covering software testing agents, development agents, operations agents, and more — with completion expected by Q1 2027.

Market explosion. Analysys's "China Office Agent Platform Market Report 2026" shows that desktop-native AI office agent applications saw explosive growth starting in March 2026. Combined monthly visits for top products surpassed 20 million, and the top five in growth rate all exceeded 100% month-over-month. Eight of the top ten products are Chinese-made, with Tencent, ByteDance, and Alibaba-affiliated products each exceeding 12 million monthly visits.

Industrial deployment. LingShu Intelligence, the first company in Shanghai to receive regulatory filing for an industrial vertical LLM, saw 2025 profitability and 2026 explosive growth — its H1 2026 revenue already surpassed the entirety of 2025. One case study says it all: at a manufacturing client, blueprint design speed went from "two days per drawing" to "20 minutes per drawing."

Capital flows in. Since June, two "token factories" — PAI-O Cloud and SiliconFlow — have filed for IPOs in Hong Kong, with agent-related infrastructure listed as the primary use of proceeds. Gu Chenhao, chief analyst at Huizheng Finance, noted that token consumption growth has long-term sustainability, giving related sectors high certainty for sustained growth.

Beneath the Hype, Real Obstacles

Every industry in its breakout phase deserves a dose of sobriety.

First, the integration problem. The Salesforce report notes that the average enterprise manages 957 applications, with an application connectivity rate of just 27%. A full 82% of IT leaders cite data integration as their biggest challenge, and 97% encountered obstacles during agent transformation. An agent can only "act" if it can connect, read, and operate — and right now, that foundation is far from ready.

Second, security. IDC Research Manager Wang Yiting pointed out that agent security has become critical for enterprise adoption, but China's agent security market and technologies remain nascent. Product standards, testing mechanisms, and certification frameworks still need significant development.

Third, the talent gap. Reports from Tianji Network note that a shortage of AI professionals remains one of the biggest bottlenecks for agent deployment.

None of these issues will stop the agent industry from growing. But they will determine who runs the furthest.

A Final Thought

Looking back at the last three WAICs: in 2024, the conversation was "can LLMs be useful?" In 2025, it was "which model is better?" In 2026, the question became "can AI actually get a real job done?"

That shift, in itself, is the biggest step forward.

As outlined in the government's implementation opinions — which identified 19 priority domains for agent development, spanning scientific research, industry, consumer growth, and public welfare — the future of agents lies not in how clever they are, but in how much human workload they can shoulder.

Whether 2026 will be remembered as the "Year of the Agent" is for history to decide. But one thing is certain: when AI stops chatting and starts working, the game has genuinely changed.