Author | Liu Jiayi
Cockpit AI is entering an awkward phase.
Nearly all automakers can demonstrate "ordering coffee with one sentence," but the high-frequency services users actually use daily remain limited to navigation, music, and a few other fixed functions. Large models and Agents have been integrated into vehicles, yet they are far from being truly user-friendly or consistently utilized.
Enterprise AI faces a similar dilemma. Office AI has boosted individual productivity, but achieving organizational efficiency requires overcoming three hurdles: scenario selection, engineering governance, and ROI. While automakers have deployed numerous intelligent agents, few have genuinely integrated into business processes across R&D, production, supply, sales, and service.
These two industry pain points serve as the entry point for understanding Tencent's latest moves in smart mobility.
At the Smart Mobility Session of the 2026 Tencent Global Digital Ecosystem Conference, Tencent released its next-generation smart cockpit TAI Agent, along with a suite of products including AI organizational efficiency solutions, aiming to drive AI from product features to practical application.
In Tencent's view, the competition in cockpit AI should shift from model capabilities to task completion capabilities. Enterprise AI evaluation should not merely count deployed agents but focus on whether they can truly enter business workflows.
Therefore, while technical capability is important, enabling AI to connect with real businesses, run stably, and generate value is crucial. This requires not only large models but also underlying capabilities such as cloud, maps, data loops, and service ecosystems—features that may seem less glamorous but determine implementation quality.
01 Users Need an AI Cockpit That Can Get Things Done
Over the past year, cockpit AI players have focused on three main narratives: integrating large models into cars, upgrading voice assistants, and bringing ecosystem applications onboard.
The industry narrative assumes that whoever controls the in-car entry point can capture more user time and build stickiness.
However, do users need to spend significant time interacting with the car system? Is it necessary for the car system to replace smartphones?
If we start from true user value, the answer is likely: no.
Zhong Xuedan, Vice President of Tencent Cloud and Head of Tencent Smart Mobility, believes that in the AI era, experience cannot be simply measured by internet-style screen time. What truly matters is what problems AI solves.
This means that spending less time in the car does not indicate a poor experience. If a single command completes a task efficiently, the saved time and operations are precisely the convenience brought by AI.
This aligns with the WeChat Mini Program philosophy of "use and leave." Users open a mini program with a specific need and close it once the task is done. In the cockpit context, the key is whether AI can get the job done effectively.
Tencent's newly launched TAI Agent follows this logic, with capabilities divided into two layers.
Layer 1: Integrating Yuanbao to Connect Service Chains.
Through full-duplex end-to-end voice, Yuanbao search, and Hunyuan model capabilities, the car system gains greater potential in understanding user needs, handling complex queries, and engaging in natural interaction. Meanwhile, the TAI Cockpit Assistant, TAI Travel Service Agent, and TAI Content & Entertainment Agent extend these capabilities to specific in-car scenarios.
For example, when a user states their travel preferences, the AI can recommend dining, entertainment, and leisure options along the route and adjust the itinerary based on needs. Once arrangements are confirmed, the TAI Travel Service Agent can link hotel bookings, ticket purchases, and even make restaurant reservations via AI outbound calls, helping users complete the entire process from planning to booking.
You can see that models provide understanding, agents handle task execution, and service ecosystems fulfill specific demands. The smoother the connection between these links, the less manual intervention required from the user.
Layer 2: Linking with Xiaowei to Ensure Seamless Service Inside and Outside the Car.
Another often overlooked issue in travel services is that user needs span the entire journey, not just inside the vehicle.
For instance, before departure, users check routes and plan trips; during driving, they need navigation and roadside services; after arrival, they may look for restaurants or visit scenic spots. If every stage requires reopening apps and repeating requests, the experience remains fragmented.
Therefore, the car system and smartphone should not be viewed as a zero-sum game. Tencent believes both will coexist for a considerable period.
The car system and smartphone each have applicable scenarios; the key lies in smooth service handover between the two ends. This is the significance of linking the mobile-side Xiaowei, WeChat Mini Programs, and the car-side Yuanbao.
Users can plan itineraries on their phones before departure; upon entering the car, routes and schedules sync to the vehicle system; after arriving at the destination, services like restaurant ordering and scenic spot guides seamlessly return to the phone. From pre-trip planning to in-car navigation and destination services, all aspects of travel become coherent.
Behind this lies support from Tencent's large models and data services, Tencent Maps' route planning and location services, WeChat Mini Programs connecting dining and hotel ecosystems, and WeChat Pay facilitating transactions. What appears to be simple cross-device continuity actually relies on the synergy of cloud, maps, data, and service ecosystems—a long-term advantage accumulated by Tencent.
Tencent indeed recognizes a key value of the AI cockpit: rather than keeping users glued to the car system, AI naturally integrates into the travel workflow, addressing needs exactly when required.
After all, no one intentionally chats with the large model in the car. But if you can order coffee conveniently when in a hurry, avoid searching for payment pages when parking, and continue checking your itinerary after getting out, AI naturally becomes part of daily travel.
02 Implementing AI in Organizations is Harder Than Imagined
Automakers want AI in cars to be anthropomorphic and intelligent, but their internal R&D, production, and management processes may not yet be ready to collaborate with AI.
AI in the car must understand users; AI within the automaker must understand business.
The former must bridge the gap from "hearing" to "doing," while the latter must overcome the chasm from "creating demos" to "stable implementation." This hurdle is equally difficult to cross.
Although individual productivity gains have allowed many employees to benefit from AI, individual ease of use does not equate to enterprise-wide business improvement through AI.
The most direct problem is the discrepancy between DEMOs and actual production workflows. In demonstration environments, agents often quickly complete tasks, but once entering real business contexts, they encounter issues such as incomplete knowledge, unstable data quality, complex system interfaces, and difficulty covering exceptions.
More tricky is that many problems do not stem from the model itself.
First, business requirements may be unclear, leading AI projects to select the wrong scenarios from the outset. If goals are ambiguous, even powerful models may merely speed up existing processes without improving final outcomes.
Second, data, knowledge, and business processes are fragmented. Enterprises accumulate vast amounts of business experience, but it is scattered across different departments and systems.
A new EV manufacturer mentioned in a roundtable forum that although departments have established their own knowledge systems, breaking down silos at the group level, managing permissions, and allowing knowledge to flow through technical and product foundations remains very difficult, with mature off-the-shelf solutions lacking in the market.
Organizational mechanisms are also a barrier. For example, if organizational and management styles remain stuck in the traditional digital era, building digital platforms often leads to detours. Additionally, rapid model iteration often outpaces engineering investment and risk governance.
These compounded issues mean that as AI delves deeper into business, automakers need not just more tools, but a systematic mechanism capable of continuously discovering needs, organizing collaboration, accumulating experience, and controlling risks.
Tencent attempts to advance this through three levels: tools, platforms, and business co-creation.
WorkBuddy and CodeBuddy target individual employee efficiency, helping staff quickly complete tasks and explore use cases; the Enterprise Intelligent Agent Platform targets business processes requiring long-term operation and multi-person collaboration.
The former makes it easier for employees to adopt AI, while the latter hosts verified Skills and Agents, allowing effective experiences to be reused by other teams and managing task scheduling, resources, and security through platform mechanisms.
These two capability layers should not be conflated. Providing only tools to employees may boost individual efficiency but still leaves old problems like redundant development and system fragmentation. Only by precipitating verified capabilities into assets can individual exploration transform into sustainable organizational resources.
However, platforms are not panaceas. Unclear business requirements and poor data quality will still cause agents to fail in real environments.
Therefore, Tencent uses the FDE (Forward Deployed Engineer) mechanism to dive deep into business sites, working with technical personnel, business consultants, and automaker teams to identify high-value scenarios, clarify data conditions, human-machine collaboration boundaries, risk controls, and evaluation metrics, then pushing applications from Demo to actual business. The "Rocket Class" co-creation between Tencent and Changan Automobile is a practice of this approach.
Tencent is also exploring using agents for vehicle planning, R&D data analysis, production-sales coordination, and dynamic marketing scenarios. Reportedly, a vehicle planning assistant for a certain automaker group shortened the data analysis cycle from weekly to hourly, increasing survey report generation efficiency by 95%.
Such data should be interpreted in the context of specific tasks and statistical standards, but it at least indicates that AI is beginning to enter business segments with clear goals and evaluation methods, moving beyond mere technical demonstrations.
03 Why Does Tencent, Which Doesn't Build Cars, Handle All AI Tasks for Automakers?
C-end users perceive changes in AI cockpits and smart driving products, while B-end automakers see AI-driven organizational efficiency. However, Tencent's presence extends far beyond these visible domains. Beneath the surface, Tencent's entire AI capabilities have been consistently feeding back into the productivity evolution of the automotive industry.
It is worth noting that AI is an extension of Tencent's existing capabilities, not a new ticket for entry.
In 2017, Tencent entered the automotive industry, initially serving automakers' data processing and smart driving R&D through cloud computing and map capabilities;
In 2018, Tencent proposed the "Smart Co-pilot" concept for cockpits. TAI subsequently iterated continuously, evolving from an initial cockpit solution to a service system centered on intelligent agents.
According to Tencent, it currently serves over 100 automakers and mobility technology companies. Its cockpit solutions have achieved mass production deployment in brands like Changan, NIO, and Leapmotor, with cumulative installations exceeding 20 million units.
These accumulations are releasing new value in the AI era. As Tencent upgrades its models, agents, and engineering capabilities, its automotive business expands synchronously. Tencent no longer focuses solely on a single product but digs vertically and extends horizontally along the automotive industry chain.
Vertically, it continues to deepen along the technological chain of automotive intelligence.
Beyond smart cockpit solutions, in the smart driving domain, Tencent Cloud continues to provide computing power, data processing, model training, and simulation replay capabilities, helping automakers improve R&D efficiency. As Chinese automakers accelerate overseas expansion, Tencent extends cloud infrastructure, smart driving maps, and cross-border data compliance capabilities to international markets.
Horizontally, it moves from in-car products to automaker organizations and further into more intelligent scenarios.
Tools like WorkBuddy and CodeBuddy help automaker employees improve personal efficiency, while the Enterprise Intelligent Agent Platform attempts to precipitate effective experiences into reusable organizational capabilities.
Meanwhile, Tencent extends its capabilities to embodied intelligence and Robotaxi: the former reuses experience in computing power, models, and data loops from the smart driving field to explore scenarios in auto factories and warehouse logistics; the latter combines cloud, maps, and mobility ecosystems to help Robotaxi companies advance commercial operations.
From inside the car to outside, from R&D to marketing, and into new intelligent scenarios, Tencent is expanding the scope of business it participates in.
However, vertical and horizontal expansions are not parallel lines; their intersection returns to the cloud and AI foundation, data loops, and Tencent's long-accumulated content and service ecosystems. Only by combining these capabilities can Tencent solve problems across an entire business chain.
But this is also where the boundary of cooperation between Tencent and automakers lies.
Zhong Xiangping, Vice President of Tencent Group and President of Tencent Smart Mobility, emphasized at the conference: Automakers are responsible for building good cars, while Tencent is responsible for making AI usable and achieving better results.
In other words, Tencent insists on not building cars and does not need to make R&D decisions or define whole-vehicle products for automakers. What it aims to do is lower the threshold for automakers to apply digitalization and AI technologies, allowing AI to truly play a role.
Just as in the past, when choosing cloud service providers, automakers had to weigh the investment in self-built infrastructure, global deployment speed, system maintenance, and compliance costs.
In the AI era, these trade-offs have not disappeared but have become even more complex. Automakers must consider computing power demands for model training and inference, engineering deployment and security management for enterprise agents, and infrastructure and data compliance requirements for overseas businesses in different markets.
Self-building is no longer an absolute path to Rome. It represents higher costs, unnecessary pitfalls, and detours.
Against the backdrop of fully embracing AI, automakers should think more calmly about which links can be completed through external partnerships to achieve the fastest and sustained results.
Perhaps concentrating resources on whole-vehicle products, core technologies, and user experience, while delegating some generic but complex digital capabilities to mature partners, is a more ideal choice.
AI does not automatically become productive force simply by entering the automotive industry. The tough battle for automakers in the next stage is not about who adopts AI first, but who can truly embed it into products and business processes and establish capabilities for sustainable iteration.