In-car AI at Beijing Auto Show: Progress and Challenges | Andrew Hart posted on the topic | LinkedIn

Executive Summary

CEO at SBD Automotive Report this post You’ve probably seen a lot of AI headlines coming out of the Beijing Auto Show. But what does in-car AI actually look like in practice? And more importantly, is it solving real problems or just searching for them? Last time I visited China, in-car AI use cases felt quite weak (“Create a desktop image of a cat eating a hotdog”), and the user experiences were often clunky. This year felt different. There’s been a clear step change in functionality, integration, and overall experience. Our team tested all the latest systems - here are some of the most interesting shifts: 🚗 Cars are starting to understand the world around them “What’s that building over there?” → instant, contextual answers about your surroundings 🧠 You no longer need to think in steps “Take me to the best Indian restaurant, and pick up flowers on the way” → The car interprets intent, suggests options, and plans multi-stop journeys dynamically ⚠️ The interface is shifting from information-heavy to light queries “What’s that icon on the cluster?” → clear, natural-language explanations of vehicle status 🎯 The car is adapting to how you feel “I’m feeling tired” → automatic adjustment of temperature, seating, and environment 🤝 The car is starting to feel less like a tool and more like a companion “I see you’re wearing a different jacket today” → subtle, human-like awareness of passengers and context Getting all this right is hard. It requires a mastery of sensors/actuator integration, app/cloud/marketplace connectivity, advanced memory, onboard processing, careful HMI design, data privacy guardrails, validation testing, etc. Although our team encountered plenty of unfinished experiences that hadn’t yet nailed some of the above, the progress is undeniable (IM Motors stood out in particular). Which brings us to the big question: Is technical capability progressing too far ahead of commercial maturity? None of the OEMs we spoke to had a strong id