WAIC 2026: When AI Stopped Showing Off and Started Getting Things Done

On July 20, the four-day 2026 World AI Conference (WAIC) wrapped up in Shanghai. Across more than 100,000 square meters of exhibition space, over 1,100 companies showcased 3,000-plus exhibits, with 351 products making their global debut.

But beneath the headline numbers, a more significant shift was unfolding: the conversation has moved from 「how smart is the model」 to 「can AI actually get things done.」

Agents: From Concept to Production

Agent was the undisputed keyword of this year's conference. Agent-related content dominated over a hundred forums, and multiple 「Crown Jewels」 award winners fell into the agent application category.

A senior official from China's Ministry of Industry and Information Technology noted that agents are becoming a key driver of new productive forces, with the industry transitioning from isolated pilots to scaled deployment.

StepFun's STEPX Neo agent phone was a standout demo. A staff member simply input travel requirements, and the system autonomously planned an itinerary, booked hotels, arranged transportation, and handled payments — all without further human intervention. SenseTime launched SenseNova U1 Pro, positioning it as a native multimodal agent foundation model purpose-built for long-horizon task execution.

Moonshot AI's open-source K3 model, released just before the conference, also emphasized specialized optimization for agent scenarios. Parameters are still growing — but they are no longer the whole story.

Robots Step Off the Stage

Embodied intelligence became a core WAIC track for the first time this year. Unlike previous years where robots danced and did backflips, companies brought real work scenarios directly into the exhibition hall.

Zhiyuan Robotics demonstrated warehouse logistics deployment, simulating actual production workflows. Keenon Robotics recreated commercial environments like coffee kiosks and retail stores on-site. Sichuan Embodied Technology's humanoid robot 「Aiqiu」 — which had already completed multiple rounds of real-world validation in Chengdu — handled guided tours and Q&A sessions throughout the venue.

The industry is undergoing a critical transition: robots are no longer showing what they can do, but what they actually do. The march from lab to production line has become a shared priority.

The Infrastructure Race Behind the Scenes

While robots and agents occupied the front stage, the real battle was unfolding at the infrastructure level.

Infinigence unveiled its 「Front Store, Back Factory, One Center」 Agentic Infra strategy at the conference, aiming to unify heterogeneous compute scheduling and boost token production efficiency. Its cross-region training system now spans multiple mainstream chip architectures and has completed cross-regional hybrid training in practice. A unified compute grid platform, built on research from the Shanghai AI Laboratory, was also officially launched.

A few years ago, the question was: do we have enough chips and compute? Today, the focus has shifted to organizing existing resources more efficiently.

Startups: Hunting for Real Markets

The FutureTech venture zone was packed with early-stage teams. Unlike previous years, when most startups clustered around foundation models, this year's cohort is pivoting toward verticals — healthcare, legal, office productivity, manufacturing.

Baixing Intelligence launched OntoZ, an enterprise ontology system designed to help companies build continuously evolving agent architectures. As founder and CEO Han Mei put it: 「The next phase of enterprise AI is not deploying more isolated, single-purpose agents. It's about building an enterprise ontology that understands the business, unifies agent decision-making, and evolves based on real operational outcomes.」

For many startup teams, AI is no longer a capital markets narrative — it is a business that needs to generate sustained value.

Closing Thoughts

During the conference, 29 countries signed the Agreement on Establishing the World AI Cooperation Organization. The Shanghai Institute of Physical Intelligence and Robotics was inaugurated. The 「UniAI · Connecting Shanghai」 initiative was launched.

These developments underscore a key reality: AI's progress no longer hinges solely on model breakthroughs. Infrastructure, industrial ecosystems, and real-world applications must advance in tandem.

If the past few years were about proving what AI could do, WAIC 2026 was about AI stepping up to actually do the work — in factories, warehouses, hospitals, offices, and everyday life. Its role is shifting from an answer-providing tool to a task-completing partner.

That may be the most valuable signal this year's conference left for the industry.