Beyond Model Size: What WAIC 2026 Tells Us About the Agent Economy

Walk through the exhibition halls at WAIC 2026 in Shanghai, and you will notice something different. The towering billboards that used to scream "largest parameter count" are gone. In their place: AI agents quietly getting work done.

The 2026 World Artificial Intelligence Conference, running July 17–20, brought together over 1,100 exhibitors, 140 forums, and nine Turing Award and Nobel Prize laureates. But the real story is not about scale. It is about a fundamental shift in what the AI industry considers progress.

From Chatting to Doing

For the past three years, the AI narrative revolved around building bigger models. Parameter counts, benchmark scores, and token consumption formed the primary coordinates by which we judged technological advancement.

WAIC 2026 tells a different story. Across the exhibition floor, AI is no longer presented as a chatbot. The demos feature agents that independently handle search research, code development, data analysis, and cross-application workflows. These systems are no longer question-answering tools. They understand goals, decompose tasks, invoke tools, and deliver results — autonomously.

Baidu's general-purpose agent "Baidu Dazi" was the only agent product to make the conference's Top 10 "Treasures of the Hall." Users describe their objective in natural language, and the agent handles the rest — crossing apps and files to deliver usable output. Since launch, daily queries have grown 20x.

DAA: The New Currency of AI

If one concept captures the industry consensus at WAIC 2026, it is DAA.

DAA — Daily Active Agents — was first proposed by Baidu CEO Robin Li at Create 2026 in May. During WAIC, IDC released the industry's first DAA Research Report, quantifying the global agent market:

• 2025: 28.6 million active agents globally
• 2026: projected 79.4 million
• 2030: projected explosive growth to 2.216 billion

Behind these numbers lies a fundamental shift in evaluation logic. The industry has long measured AI value by how much compute was spent and how many tokens were consumed — essentially a cost-side perspective. DAA asks a different question: how many agents are actually entering business workflows, completing tasks, and creating value every day?

Uber provides a cautionary case study. In late 2025, the company deployed AI coding tools to roughly 5,000 engineers. Within four months, they had burned through the entire annual AI budget. Uber's COO publicly stated: "There's no clear linear relationship between token consumption and valuable products shipped."

Measuring AI by tokens is like measuring a logistics company by fuel consumption. DAA reframes the question: not how much fuel you burned, but how many packages your fleet delivered.

From Software to Hardware: Agents Get Bodies

The other major thread at WAIC 2026 is embodied intelligence. Humanoid robots still draw the biggest crowds, but this year, vendors have moved past walking and dancing demos. The exhibits show machines that perceive environments, manipulate objects, and collaborate with humans in factories and warehouses.

Hu Shuang, general manager of Joyson Embodied Intelligence, put it bluntly: "The path to general AI in the physical world is not in papers or on stage. It is on factory floors, on production lines, in every real operation."

BluePoint Touch showcased force sensors that determine whether a robot can "gently pick up an egg." These seemingly tiny components are transitioning from supporting roles to leading roles in embodied intelligence deployment.

Huawei exhibited the Ascend 950 Supernode. Kunlun Core showed its Tianchi 256 super-node. These infrastructure upgrades point in the same direction: the compute demands of agents and embodied AI are shifting from training-heavy to inference-heavy. An agent that continuously reasons, responds in real time, and makes autonomous decisions demands a fundamentally different architecture than a model invoked once.

AI Governance: From Theory to Practice

Perhaps the most telling shift at WAIC 2026 is that governance has become a recurring theme — not in the abstract policy-document sense.

When AI agents autonomously navigate between software platforms, robots operate in physical spaces, and AI systems embed into healthcare and finance, questions of data security, privacy, and system reliability cease to be "future concerns." They are immediate engineering problems.

On the eve of the conference, representatives of 29 countries signed the Agreement on the Establishment of the World Artificial Intelligence Cooperation Organization in Shanghai. China committed to providing 5,000 AI training slots for developing countries over the next five years and deploying its "Mazu" meteorological early-warning system across 30 nations.

Zhen Lixin, chairman of Intsig, distilled the industry direction into three words: "upward, toward good, and innovative" — technological self-reliance, unwavering safety guardrails, and a relentless focus on real-world application.

The Bottom Line

Every major AI conference holds up a mirror to the industry. WAIC 2026 reflects an industry leaving its showcase phase and entering its deployment phase.

That does not mean models no longer matter. It means they are no longer the whole story. When Microsoft CEO Satya Nadella casually mentions "we already have 20 million AI agents running in the background," and IDC projects 2.2 billion agents by 2030, the direction is unmistakable.

The keyword for AI's next chapter is not "bigger." It is "deployed."