2026 Beijing Auto Show Spotlights In-Car AI Transformation: From "Showpiece" to Doer

At the 2026 Beijing Auto Show, numerous automakers and tech companies showcased in-vehicle intelligent agent solutions centered on "Agentic AI," marking a pivotal shift in automotive artificial intelligence—from generative AI focused primarily on conversation and entertainment toward practical agents capable of perception, decision-making, execution, and closed-loop operations. In recent years, in-car AI has largely offered limited-value functions such as voice interaction, joke-telling, or poetry composition. The new trend highlighted at this year’s auto show emphasizes that AI must solve real-world driving scenarios—such as automatically adjusting cabin environments, planning itineraries, and handling multi-user commands—transitioning from “passive response” to “proactive service.” Companies including Tencent, Volcano Engine, Lenovo, SenseTime’s Jueying, and MediaTek unveiled intelligent agent platforms based on distinct technical approaches. Volcano Engine launched an Agentic AI-based solution already deployed across 50 automotive brands and 145 vehicle models, completing over 30 million closed-loop cabin services daily. MediaTek introduced its “AI-defined vehicle” concept, with its Dimensity C-X1 cockpit platform delivering 400 TOPS of computing power to support on-device large model deployment. Lenovo rolled out its Auto AI Box and OneAI platform powered by NVIDIA DRIVE AGX Thor-Z chips, achieving millisecond-level interaction latency. Meanwhile, SenseTime’s Jueying demonstrated Sage Box for always-on, on-device responsiveness and plans to extend its agent capabilities into the Robotaxi domain. Technologically, the industry is converging around three main directions: 1) **On-device intelligence**, prioritizing local computing for speed and privacy; 2) **Ecosystem platforms**, exemplified by Tencent leveraging WeChat Mini Programs, payments, maps, and other services to enhance agents’ task-execution capabilities; and 3) **Whole-vehicle intelligence**, such as Geely’s “1+2+N” multi-agent framework, where a vehicle-level agent orchestrates systems including the cockpit, autonomous driving, and chassis. Industry consensus holds that successful deployment of intelligent agents hinges on three concurrent advancements: safe and controllable AI models, progress in engineering paradigms (e.g., Manus-style constraint mechanisms), and effective integration of in-vehicle and external service ecosystems. A cloud-edge协同 architecture is seen as the long-term path forward—on-device processing for immediate tasks and cloud resources for complex scenarios. Global automakers are also accelerating their efforts. Volkswagen announced plans to mass-produce vehicles featuring full-domain Agentic AI by 2026. BMW unveiled three China-customized intelligent agents built on Alibaba’s Qwen model. Mercedes-Benz showcased AI cockpit and driver-assistance innovations powered by its MB.OS operating system. On the policy and market front, China is intensifying support for intelligent agents—the term appeared for the first time in the Government Work Report, and the Ministry of Industry and Information Technology (MIIT) has explicitly called for accelerated product iteration. Market data shows the global automotive AI agent market was valued at $1.62 billion in 2025 and is projected to reach $13.98 billion by 2032, with China expected to account for over 35% of the total. Meanwhile, penetration of L2 advanced driver-assistance systems (ADAS) in new vehicles has already surpassed 60% and is forecast to exceed 70% in 2026. Nonetheless, challenges remain. Research institutions caution that if agents fail to achieve demand anticipation, long-term memory, and multi-task handling, they risk becoming features users ignore. The key to future competitiveness lies in establishing commercially viable, scalable, continuously evolving, and cost-controlled closed-loop solutions.