Published on Oct. 30, 2026

"The integration of AI and automobiles isn’t about adding features to the car. It’s a ‘brain transplant,’" Xiaomi founder Lei Jun said at the 2026 World Intelligent Connected Vehicles Conference in Beijing on October 21.

In the meantime, other carmakers have started bragging that they have completed the first batch of these "brain transplants." On October 23, SAIC Roewe launched the Jiayue 07, co-developed with ByteDance’s Volcano Engine. They called it the world’s first mass-produced "AI-native" car.

Rather than adding AI to an existing design, the car was built for AI from the ground up, its developers say. More than 2,000 software interfaces let AI coordinate the whole vehicle as a single agent that can listen, reason and act.

Its two "brains" are ByteDance’s Doubao cockpit assistant, which speaks 12 Chinese dialects and 22 foreign languages (we are not sure what is more impressive), and Momenta’s R7 world model for assisted driving, which responds to plain-language commands such as "I’m running late, drive faster." Priced from RMB 119,800 (around US$18,000) with launch promotions, the car allegedly had more than 51,000 pre-orders before launch.

AIVA’s ME7 pushes the same idea even further. It pairs the car with a wearable and an AI agent that carries the user’s preferences beyond the two hours a day spent driving. "We can’t just build cars," one employee told Chinese tech outlet Wandian (in Chinese).

Building AI infrastructure is expensive, slow and out of reach for most smaller automakers. On October 23, at the World Intelligent Connected Vehicles Conference, the China Mobile-led National AI Application Pilot Base for the automotive sector unveiled what it calls the industry’s first “automotive AI foundation platform.”

The platform bundles computing power, data and models into a shared service. It manages 112.7 EFLOPS of compute, lets carmakers keep ownership of their core data, and offers a full toolchain for developing and deploying models. Six AI agents, spanning design, process planning, quality inspection, maintenance and autonomous-driving development, are already in use at more than ten state-owned carmakers.

The base reports first efficiency gains. In Shanghai, the process planning agent can generate process documentation for a vehicle model in just four hours instead of five days. In Jilin province, the visual quality inspection agent has increased defect detection rates to more than 99.5%, while reducing the time to adapt the system to a production line from four weeks to one week.

One automaker executive at the launch said the platform "effectively equips the entire industry with computing power, data and models in one go," so that carmakers can stop duplicating infrastructure spending and focus on products and core technology. For China’s state telecom giants, the launch marks a shift from selling raw compute to packaging industry-specific AI agents.

Chip design has long been limited by scarce engineers and long working hours. Empyrean Technology, China’s leading domestic EDA (electronic design automation) vendor, wants to change that. At its HUG 2026 user conference, the company unveiled its .aix series of AI agent platforms, built on Chinese open-source large models.

By pairing each EDA tool with a Copilot agent, the platform “upgrades traditional tools into intelligent experts that work 24/7, so chip design workflows are no longer interrupted by the rhythm of human work,” the tech outlet Kuai Keji (in Chinese) reported. The company describes its approach as"Design by AI, verification by sign-off." Agents generate and iterate on designs, while Empyrean’s established verification engines provide the final check.

For analog circuits, parameter tuning and layout that once took weeks or months now take less than a day, with 60% fewer design iterations, according to Empyrian. Display-panel pixel design that took an expert a week now fits into a single day. In an open-source CPU project, work that would normally occupy dozens of engineers for two to three months was finished in a few days.

The strategic aim was to help Chinese analog-chip makers close the gap with overseas rivals in the breadth of their product portfolios, Kuai Keji wrote. Of course, those foreigners might have access to similar tools.
The launch came the same week Cadence released its own in-house-deployable chip-design AI models, a sign that agentic EDA has become a global race that China intends to compete in.

Instead of developing a "super AI doctor," Tairex, a startup incubated by Tsinghua University’s Institute for AI Industry Research, builds physician digital-twin agents that learn individual clinicians' diagnostic logic.

On October 21, 2026, it launched a multidisciplinary human-agent collaboration ecosystem built on its "Agent Hospital" technology, which Nature featured earlier this year. The system’s 54 AI doctor agents, spanning 27 clinical departments, first train in a virtual hospital full of simulated patients and score over 96% on the MedQA benchmark. They are then deployed alongside real physicians and gradually adapt to each doctor’s reasoning and habits.

Nearly 200 doctors across 47 departments in nine pilot hospitals, from top-tier urban hospitals to community clinics in western border regions, now have their own AI twins. For complex cases, specialist agents consult each other in real time. In one example, a respiratory agent flagged that a drug proposed by an endocrinology agent could endanger an asthmatic patient.

Safety remains the hard part. Tairex’s Chief Product Officer Qiao Yuchen told Kechuangban Ribao (in Chinese). that the bar varies by task. "Some decisions are extremely critical, such as diagnosis. Once a misdiagnosis occurs, the entire subsequent process goes off course." Follow-up care and routine consultations, he added, leave relatively more room for error.