Author | Liu Jiayi
In the broader context of Physical AI, the urgency of building an AI-powered car and transforming into an AI-driven enterprise is equally high.
However, when it comes to transformation, most automakers, even after implementing sweeping organizational reforms, still find themselves at a loss: where exactly should they focus their efforts for AI to become the "game-changing move"?
Discussing this abstract issue is complicated. It is better to take a more practical approach and answer a more realistic question: How can AI achieve quality improvement and efficiency gains?
Today, efficiency is enough to determine an automaker's market position. The iteration pace of R&D models, the scheduling capability for capacity ramp-up, and the early warning and adjustment of the supply chain—slowing down in any of these means falling behind.
One solution is to build an AI system internally. For instance, BMW has built a top-down agent system called "Gaia." Another approach is to find a suitable AI platform for deep co-creation.
The collaboration between Leapmotor and WorkBuddy has successfully validated this logic.
Today, WorkBuddy announced its official "onboarding" at Leapmotor. AI Agents have been deployed in over 40 core business scenarios, spanning complete vehicle R&D and manufacturing as well as procurement supply chains, covering core business domains such as software quality, engineering digitalization, intelligent cockpit evaluation, and laboratory management, achieving an average efficiency improvement of 90%.
Two conclusions can be drawn from this: First, existing intelligent platforms have gradually reached the maturity required to enter the core business domains of automakers, with proven implementation results. Second, AI is no longer just an independently invoked programming tool; it integrates into the organizational context in a more intuitive and fundamental way, becoming a core productive force for automakers.
01: "Small but Painful" Needs Are Solved Immediately
When Tesla was deeply mired in a capacity crisis, Elon Musk pointed out a "misalignment" incisively. Design engineers sat in offices far from the production line drawing blueprints, while the people who actually had to turn those blueprints into cars were repeatedly making adjustments through trial and error on the factory floor.
Efficiency is essentially a "density" problem. Once the density gap between personnel becomes too large, information and instructions will continuously degrade during transmission.
The solution is to bridge this gap, just as Musk eventually moved the engineering team's workstations right next to the assembly line, ensuring the design side considered production reliability from the very beginning, allowing problems to be fed back and adjusted on the spot.
However, in the massive organization of an automaker, which hosts thousands of employees and hundreds of process systems, such gaps are almost ubiquitous and hard to eliminate entirely. Major projects prioritize filling the main arteries, while those "small but painful" needs are becoming the bottleneck dragging down efficiency.
A typical case is stamping die management.
As a core process in complete vehicle manufacturing, the precision of stamping directly determines quality. A large set of dies can weigh tens of tons, with a long cycle from design to manufacturing, and the management process is extremely tedious. Die administrators have to check manual ledgers every day, compare the cumulative stamping times for each die, and estimate which ones are nearing their end of life and which ones need maintenance.
This is not even a matter of simply replacing them on schedule. Different dies have different wear patterns, and the wear rate of the same die varies completely under different stamping materials and machine parameters. Accurate judgment relies purely on experience.
Ideally, a software system covering 12 modules, including die ledgers, stamping time records, maintenance and repair, life monitoring, and scrapping judgment, should be built.
However, under the traditional process, the requirements and design phase alone would take 12 person-days, system development about 70 person-days, totaling approximately 100 person-days. Moreover, on the IT schedule, projects like MES upgrades, ERP transformations, and data middle platform construction, which require hundreds of person-days and impact the entire company, occupy the vast majority of IT capacity. Needs like stamping die management, which involve "one workshop, one module," are not strategic enough, and their ROI is not impressive enough, so they can only be pushed down the priority list.
Ultimately, the more such needs are backlogged, the greater the communication loss between IT and business, and the lower the delivery efficiency.
The same logic applies to the test vehicle roll-off and delivery level. Before a new car goes on the market, an Excel file is passed back and forth among several departments, with feedback information omitted and wrong versions sent out from time to time.
These problems are common ailments for all automakers, but now, Leapmotor, in collaboration with WorkBuddy, is solving them first.
On the one hand, non-IT personnel can independently resolve development needs without waiting for scheduling. For example, developers built a stamping die management platform themselves, taking 10 person-days and going live in two weeks, saving about 90% in investment. For the first time, full lifecycle management of dies has a unified, real-time, and traceable digital platform.
On the other hand, after the efficient implementation of AI applications, personnel collaboration and communication efficiency have significantly increased. Similarly, the test vehicle roll-off and delivery tracking system built by employees themselves covers the coordination of three links: roll-off, temporary license plates, and transportation, achieving real-time tracking and timely feedback for the status of 69 test vehicles. Under the traditional process, development would have taken 45 person-days; this time, it was completed in just 5 person-days, also saving about 89%.
Essentially, WorkBuddy has disrupted the traditional development process. The past cross-departmental chain of "requirement alignment—scheduling wait—development delivery" has been compressed into internal operation and follow-up resolution within the requiring department, significantly reducing alignment costs.
The efficient construction of digital platforms also makes data circulation more convenient and information sharing more timely. When the "gaps" between departments and personnel are tightened, the gears of the entire automaker's efficiency naturally turn faster.
02: The Know-How of Veteran Workers Can Be Turned into Skills
Within the Leapmotor system, there are already over 40 cases similar to stamping die management and test vehicle roll-off and delivery.
In the complete vehicle business line, the software quality management platform, intelligent cockpit evaluation system, complete vehicle manufacturing digital middle platform, and laboratory information management system have all achieved tangible quality improvement and efficiency gains with the help of WorkBuddy.
In addition, WorkBuddy is also being rolled out simultaneously in electronic product lines such as connected vehicles and intelligent driving, as well as key links like supply chain procurement and auditing. For example, it helps quickly build prototypes and accelerate iterations for intelligent driving navigation products.
It is no exaggeration to say that AI platforms represented by WorkBuddy are seamlessly connecting the core operational arteries of automakers, allowing business flows, data flows, and collaboration flows to be fully integrated from now on.
From this, it can be seen that the onboarding of WorkBuddy is essentially the introduction of three capabilities: AI Coding, Agent collaboration, and the Skills ecosystem.
This means it is simultaneously doing three things:
- Enabling non-IT personnel to independently resolve development needs; ensuring the efficient implementation of AI applications; and distilling business experience into reusable platform capabilities.
Behind the digital efficiency gains is the consolidation of experience. Encapsulating the Know-How of veteran workers into Skills is a matter with a very strong "long-tail effect." It represents the "key" for super individuals to evolve into super systems, and then into super teams. Just like in the die development system, the system can summarize the experience of senior employees, transform it into executable operating procedures, and ensure that judgment criteria no longer rely on a single individual.
In the procurement and supply chain risk control links, the experience of senior procurement personnel in judging component price fluctuations and the logic for identifying abnormal operational risks of suppliers are broken down and encapsulated into skills.
The same applies to the pre-event health check expert for connected vehicles. Before the traffic peak during holidays, it automatically conducts a comprehensive "health check" on the multi-cloud environment, replacing the tedious manual inspection of logging into each cloud and checking items one by one, turning "veterans staring at screens" into "automated system runs."
Therefore, encapsulating Skills is not about simply and crudely replacing "human" capabilities, but rather integrating the experience scattered across individuals, giving the system the foundation for continuous evolution. For example, proactively warning of supplier risks and sending alerts before abnormal signals reach the production line; automatically discovering hidden dangers in the connected vehicle system, intercepting problems in the process in advance that would otherwise require post-incident reviews to be found.
When Skills are connected and call each other within an ecosystem, the system no longer just waits for humans to feed it experience. It begins to discover patterns, recommend combinations, and optimize processes on its own, proactively participating in the operation of business flows.
This evolution, in turn, feeds back to the organization and business personnel. Essentially, Leapmotor's more than 40 AI applications are not developed in isolation but grow within a unified Skills ecosystem. Every time a new Skill for a business scenario is added, the capability boundary of the entire platform expands once. Individuals contribute Skills, Skills enhance the platform, and the platform feeds back to individuals. Once the flywheel starts turning, AI Native truly transitions from a concept into the organizational fabric.
03: Coding Is Just the Foundation; AI Must Enter the "Organizational Context"
Currently, the industry's imagination of AI mostly stays on office tasks like "writing code faster" or "making a PPT with one sentence." Because the direct perception is the most intuitive, when an engineer uses AI, the code output efficiency visibly improves.
But this level of capability is precisely the most fragile.
Model iteration speed is measured in months. Today, a certain tool leads in code generation, and tomorrow it might be caught up with or even surpassed by a stronger model. AI Coding is essentially a thin layer of interface.
Moreover, the individual efficiency gains brought by AI programming have not significantly translated into throughput improvements at the organizational level, and have even led to a situation where the more AI is used, the busier people become.
The three hours saved by engineers using AI for programming are often re-spent on acceptance, rework, and cross-departmental alignment.
In the computer science field, there is Amdahl's Law, which states that the overall speedup of a system is always limited by the proportion of the system that is not sped up.
Comparing a system to an organization, within any organization, collaboration is what directly targets the core of efficiency. Therefore, if an AI tool only improves the speed of a single node, its effect is merely superficial. Only by integrating the organizational context and allowing AI to truly understand the rules and systems of how each part operates can it grow into a digital employee that can get up to speed quickly and collaborate efficiently.
The entry of WorkBuddy is exactly like this, minimizing the "parts that are not sped up" to the extreme.
Business personnel programming themselves bridges the process. A requirement from proposal to implementation no longer needs to be translated back and forth across departments. Similarly, encapsulating expert experience into Skills hands over the knowledge context.
Going further down, what AI absorbs is the organizational context itself. The implicit rules originally scattered in departmental silos, Excel sheets, and the minds of veteran employees are loaded into a platform for circulation and value addition.
Essentially, the capabilities of WorkBuddy have taken root downwards, growing into a part of the enterprise's organizational capabilities and evolving into non-transferable digital assets for the enterprise.
From another perspective, why is Leapmotor, as a benchmark among new forces, so eager to push this forward?
Leapmotor's current core strategy is full-domain in-house R&D and component innovation. The underlying logic of this approach in the past was to use engineering capabilities, such as self-built factories and in-house component R&D, to push down costs and boost efficiency.
The entry of WorkBuddy takes it a step further. With AI capabilities as the foundation, it extends this engineering efficiency gene from the manufacturing end to every business link.
Other automakers cannot escape this same challenge.
Leapmotor is not an isolated case. Changan Automobile and Tencent jointly launched the first "AI Rocket Class," selecting business and technical backbones internally. Using WorkBuddy and CodeBuddy as practical training platforms, it covers full-link training and practice in scenario decomposition, knowledge base construction, Skill package development, and multi-Agent orchestration. The goal is to train these individuals into roles that understand both business and can build AI applications hands-on, creating "silicon-based employees" that can go online and be operated.
This also returns to a more fundamental judgment. For automakers to transform into AI Native organizations, they do not need to wait until AI capabilities are perfectly mature before taking action, nor do they need to complete a thorough organizational transformation before introducing AI tools.
Solving actual pain points first, letting business personnel use it first, allowing experience to consolidate first, and letting the organization naturally evolve through the use of AI capabilities—this is the transformation pace that better fits the realistic conditions of most automakers.
After all, AI Native organizations emphasize "Native." An organization embracing AI on its own will go much further and deeper than AI being forcibly implanted into the organization.