He Xiaopeng Talks about Organizational Evolution Again: Every Department Should Use AI

Text | Chen Xi

At the beginning of this year, He Xiaopeng, the CEO of XPeng Motors, conducted an internal review. He pointed out that all departments should embrace AI.

When the new car P7+ was launched in November last year, XPeng first proposed the concept of "AI cars". However, some employees said that He Xiaopeng's attention to AI is not limited to car products and large models for assisted driving. He also studies which of the more than 30 first - level central departments within the company have embraced AI and how much knowledge employees have learned through AI.

"AI can make a larger and stronger enterprise more powerful in terms of group efficiency and group compound interest." He Xiaopeng recently emphasized the role of AI when talking about organizational capabilities at the Feishu Future Unlimited Conference. He hopes to bring out the value of AI in the management of more than 20,000 employees globally.

Currently, the competition in the Chinese automotive market is intensifying. The Xiaomi YU7 received 289,000 large orders within an hour of its launch. With high - intensity order explosions and brand waves sweeping across, Chinese automakers have to think deeply about how to run fast to win more opportunities.

After experiencing disputes over product configurations, organizational changes, and gradually emerging from the shadows in the past two years, the leader of XPeng Motors has realized that in the ten - year growth of new - energy vehicle startups, the victories achieved by relying on a single blockbuster product and local advantages in the past cannot last. More blockbuster products and stronger anti - risk capabilities of the enterprise come from continuous evolution of organizational management. He believes that if a company can do well in the "technology - group - compound interest effect", it is more likely to succeed.

He Xiaopeng has publicly stated on many occasions that the era of AI cars has arrived. XPeng not only makes AI cars but also aims to integrate AI with travel products such as flying cars and robots.

A radical strategy not only requires perseverance but also a system. In the era of fuel - powered cars, Toyota Motors implemented extreme management of parts inventory and produced on demand to reduce costs. "TPS lean production" has become a model for global automakers to survive.

But in the AI era, how should automakers build a survival system exclusive to AI cars? This is an important topic for He Xiaopeng this year and perhaps a common challenge for automakers.

Turning the tables, XPeng's survival philosophy has changed

On July 4th, XPeng Motors launched its new car G7 with a computing power of over 2000 Tops. Before the release, Chen Yonghai, the vice - president of XPeng, said on the Weibo platform that it was "impossible" for the G7, which is equipped with core configurations such as an 800V high - voltage platform, a range of over 700 kilometers, more than 2000 TOPS of computing power, standard CDC, and a high - quality AR - HUD, to be priced below 200,000 yuan.

However, when it was finally released, XPeng added a model version with a range of 602 kilometers, and the price dropped to 195,800 yuan.

After all, in the face of the phenomenon - level blockbuster Xiaomi YU7 and the ever - popular Tesla Model Y, it is much more difficult for automakers' new pure - electric SUVs to break through. With its high computing power and other configurations, combined with the price - cut strategy, the G7 received over 10,000 large orders within 9 minutes of its launch.

This is a continuation of XPeng's "small victories". Since the end of last year, XPeng has gradually turned the tables: the MONA M03 has sold over 15,000 units for several consecutive months. The P7+, new G6/G9, and the new G7 contributed over 190,000 sales in the first half of this year for XPeng, which exceeded the total sales in 2024.

Within this year, XPeng will also launch models such as the new P7, and its layout is getting closer to that of mature automakers.

He Xiaopeng also summarized in an interview after the recent G7 press conference that from the end of 2022 to the fourth quarter of 2025, XPeng has undergone a nirvana - like and transformative change in three years. This includes recruiting "Iron Lady" Wang Fengying from Great Wall, He Xiaopeng personally managing the supply chain, laying off 10 senior executives, re - sorting out product positioning, expanding the price range from 200,000 - 500,000 yuan to 100,000 - 500,000 yuan, going overseas, and exporting core technologies at the electronic and electrical architecture level to Volkswagen.

"In the past, we adjusted the organization, strategy, planning, operation, and supply chain. Basically, we have made many adjustments to everything you can think of that a company needs to do."

The maturity of the product matrix is an obvious change in XPeng's survival philosophy, but the more long - term change is hidden in the management of the large organization of 20,000 people.

To make the organization more efficient, He Xiaopeng has required the team to improve horizontal and vertical efficiency since 2023. When necessary, he even personally breaks down departmental barriers and establishes horizontal processes.

He advocates internally and calls on employees not to use PPTs in meetings but to use cloud documents instead. "Because the higher - level we are, the more we should realize that most people's work actually involves preparing documents, having meetings, writing meeting minutes, preparing the next document, and preparing for the next meeting."

At first, it was just a suggestion, but later He Xiaopeng "forced" senior executives to require a review of how many employees had generated, read, and annotated cloud documents. According to internal statistics in 2024, unnecessary offline meetings decreased by 30%, and a large number of emails, chats, and meetings have been shifted to cloud document collaboration.

The advantage of document - based office work is that the team's collaboration is effectively recorded and accumulated as experience. Currently, 72% of XPeng's employees use Feishu cloud documents for collaborative work every day.

Another special means to improve organizational efficiency is to shorten meeting durations. In 2024, He Xiaopeng set "shortening meeting durations" as a key OKR, saying "Try to have short meetings and talk about the essentials without wasting everyone's time." In the second quarter of this year, the proportion of short meetings with less than 8 people and less than 1 hour in XPeng reached 81.2%, compared with 74% in the first quarter.

He Xiaopeng also encourages the internal use of AI to record meeting points. In 2024, XPeng generated 160,000 Feishu meeting minutes using AI throughout the year. In the first half of this year, XPeng generated over 630,000 Feishu meeting minutes using AI, with nearly 3,500 AI - generated meeting minutes per day.

XPeng has found that the combination of AI meeting minutes and cloud documents enables rapid communication and collaboration among employees who speak different languages. Currently, XPeng Motors has more than 20,000 employees globally, with about 1,000 overseas employees from more than 40 countries. He Xiaopeng also believes that the car companies that can ultimately survive must have a scale of more than 100,000 people. How to transform from individual efficiency to large - team efficiency is very important.

From relying on the long - board of intelligent driving as the "locomotive" to trying to make up for all - around weaknesses, XPeng has been working hard to reverse the company's operating profit in the past two years. In the first quarter of this year, XPeng's revenue was 15.8 billion yuan, and the loss continued to shrink to 660 million yuan. The company is optimistic about achieving profitability in the fourth quarter.

He Xiaopeng said that XPeng will enter a period of rapid development in the fourth quarter of this year. Judging from the latest strategies, XPeng is trying to use AI to make the pace of car - making lighter and faster.

Use AI to build AI cars

Since the beginning of this year, He Xiaopeng has significantly accelerated the integration of AI and cars. He believes that AI has great potential in the physical world, and a real robot must have much stronger AI capabilities than the capabilities of its mechanical body.

On April 14th this year, XPeng officially announced its self - developed cloud - based large - model route that had been in development for half a year. Similar to the training concept of DeepSeek, which became popular at the beginning of the year, XPeng hopes to first train an ultra - large - scale base model in the cloud and then retain the model's driving ability on the vehicle side through knowledge distillation of small models.

"This can break through the model limit caused by the scarce AI computing power on the vehicle side." Li Liyun, the person in charge of XPeng's autonomous driving, previously told 36Kr Auto.

To develop the base model, XPeng has also built an AI computing power cluster with a scale of ten thousand graphics processing units and has invested over 4 billion yuan in AI in a year. At the same time, to break through the computing power limit on the vehicle side, XPeng's Turing AI chip, which was specially customized for large AI models after five years of development, has recently been mass - produced and installed in vehicles.

The new car G7 Ultra, equipped with three Turing chips, has become XPeng's first AI car with L3 - level computing power. On the G7, XPeng is trying to implement the VLA - OL model (Vision - Language - Action + Autonomous Reinforcement Learning). The official claims that the upper limit of intelligent driving ability is more than 10 times higher than that of the industry's MAX models.

The penetration of AI is not limited to car products. XPeng is also trying to penetrate AI more deeply into all business links of car production, from R & D, production, management, to operation and maintenance. He Xiaopeng said, "AI is used in nearly 200 internal business aspects."

According to He Xiaopeng's introduction at the Feishu Future Unlimited Conference recently, currently, 25% of the company's code is written by AI; AI can also empower the R & D of chassis and power systems; in the digital factory, AI vision can automatically identify the quality of products during stamping, welding, painting, and assembly processes; at the digital marketing level, AI is also used in the generation of marketing posters, etc.

AI is not only the top - priority project of XPeng but also deeply penetrates into the capillaries of organizational management.

XPeng has also jointly built an AI service desk with Feishu for the sales, after - sales, and service management of stores. Currently, in XPeng's 690 stores across the country, store employees can talk to AI, automatically retrieve relevant knowledge bases, and generate reference scripts. The proportion of manual work orders has decreased from 15% to 8%.

A Feishu employee told 36Kr that XPeng's AI service desk contains relevant knowledge in finance, human resources, administration, automotive supply chain, and quality. Against the background of the increasingly rapid iteration of new car software and hardware in the automotive industry, through AI Q & A and the underlying knowledge space, the human - resource training cost at the store level during new car and function iterations can be significantly reduced.

However, it is difficult for the daily operation of a large enterprise to be solved by a single AI dialogue system. Just as the popularity of the general AI agent product Minus at the beginning of the year was not because of its advanced technical path but because it tried to solve a problem - making the AI agent a truly useful product.

XPeng's internal digital employee IRON is a similar AI agent. It integrates 44 service desks, Feishu knowledge Q & A APIs, and XPeng's internal AI capabilities. Agile management of "Feishu Projects" was used during its development. It can not only review reimbursement forms but also screen resumes. In the first half of this year, the penetration rate of IRON in XPeng was close to 70%. "When a company truly recognizes group efficiency and group compound interest, it will undergo a huge change." He Xiaopeng said.

Feishu has made an analogy about the application of AI: "The AI introduced by an enterprise is like a fresh graduate who has strong theoretical knowledge but almost no practical experience. If the industry knowledge accumulated within the enterprise, customer leads in the CRM, and production data in the MES system are not connected, no matter how advanced the AI model is, it will always remain in the state of the first day of employment."

What's more difficult than systemization is the organization in the AI era

Since 2023, almost all companies around the world can no longer ignore AI.

From global Internet giants such as OpenAI, Google, and Meta to institutions like DeepSeek, the competition and iteration of AI technology are advancing by leaps and bounds. However, for most enterprises and teams, the more urgent task than technological competition is the implementation and application of AI.

The automotive industry is no exception. In the AI era, the strategies, products, and technological paths of automakers may change at any time. The vehicle development cycle, which used to be 4 - 5 years in the traditional fuel - powered car era, has been compressed to 18 months, and the software iteration of assisted - driving technology has evolved to a weekly basis.

The survival game of automakers requires more efficient organizational collaboration, faster information flow. More importantly, the experience and resources accumulated during continuous iteration need to be better accumulated and utilized.

After deciding to fully embrace AI, He Xiaopeng no longer regards AI as just a productive force but as an internal force of an advanced organization. "The original intention of using AI is not to replace people but to enable people to do more valuable things." He Xiaopeng said at the Feishu Future Unlimited Conference.

This perception is also penetrating the industry.

Among automotive peers, in addition to XPeng, other new - energy vehicle startups and established automakers are also looking for more efficient management methods. AI not only "manages people and affairs" but also deeply penetrates the entire supply chain, energy replenishment, R & D, and dealer management of automakers' products.

Li Auto is a typical example. After building a supply - chain management tool based on Feishu in 2021, Li Auto aggregated scattered material information and data into multi - dimensional tables. Without manual summarization, it can see the in - transit/in - stock status of parts in real - time. CEO Li Xiang also called Feishu a "super - sensing tool".

A similar case of improving organizational efficiency through efficient information integration also happened to NIO. Last year, when NIO launched its second brand, LeDao, to ensure the battery - swapping experience of the first - model car, NIO needed to complete the transformation of 1,000 battery - swapping stations within 6 months.

However, this huge task could only rely on a team of 300 people. If NIO developed a business system independently, it might take several months or even longer. Eventually, based on its internal site - selection system, the datawind dashboard function of Volcengine, and combined with Feishu multi - dimensional tables, NIO built a project management system in just one week, covering the entire process from plan setting, personnel scheduling, transformation material and battery logistics management, on - site transformation, to operation acceptance.

"Automakers can use multi - dimensional tables like building Lego to quickly digitize rapidly changing business requirements." A Feishu employee told 36Kr by analogy. Multi - dimensional tables are not just simple documents and tables but are more like a flexible, chart - visualizable, and highly collaborative database that can accommodate 1,000 people