Li Xiang on AI Tools, Automotive Strategy and Human Relationships

In April 2025, Zhang Xiaojun held an extended conversation with Li Auto founder Li Xiang, using a “CEO model” metaphor to discuss technology, strategy, organizations and AI’s relationship with people.

The host structures the conversation around technology, strategy and organization, then moves into personal memory, family, relationships and wisdom.

Li distinguishes information tools, assistive tools and production tools. He says AI value should be judged by whether it actually improves work outcomes or reduces workload, not merely produces content.

He uses navigation and driver assistance as examples of assistive tools: they help users but still require human involvement. Drivers should follow their vehicle’s instructions and remain responsible.

Using DeepSeek as an example, he discusses the relationship among research, development and market application, arguing that teams must connect frontier exploration with practical capability building.

He says organizations should not merely change tactics when problems arise; they should review results, study users and markets, then set goals and actions. This is his management approach.

He recalls the team evaluating available models and deciding to use existing capabilities as a foundation for further work, including the effect of that change on internal researchers.

Discussing vision-language-action models, he describes driving-model development as gradual, moving from combinations of perception and rules toward tighter integration of perception, decisions and actions.

He describes the team’s proposed training process, including visual and language foundations, driving data and subsequent training. Architecture and capabilities are the company’s account at the time.

He mentions training feedback on comfort, traffic rules and collisions and envisions improvement through environmental and human feedback. These plans do not establish validated driverless operation.

On agents, he imagines different business teams developing customer-service, sales or coding tools on a shared platform, rather than one team defining every specialty’s needs.

He describes the company’s engineering investment in chips, compilers and software and argues low-level optimization can help models run in vehicles. Performance claims require assessment on specific hardware and tests.

He says stronger models make safety, reliability and behavior consistent with user expectations more important, while acknowledging the challenge of interpreting and validating black-box systems.

Using new-employee training as an example, he says organizational capability also requires research, course design, real operation and improvement through feedback.

He envisions future driving products combining technical skill, reliable behavior and understanding of users. His human analogy is a product vision, not evidence that the system has a human driver’s abilities.

In the strategy section, he discusses relationships among customer demand, technology and products, organizational capability and scale, saying the company must examine them together.

He compares Apple and Microsoft historically to illustrate two choices, integrated hardware-software devices and software ecosystems, each with its own challenges.

On factories, he argues automation should focus on overall production efficiency rather than simply replacing a particular role. This is his perspective on manufacturing.

He says AI systems and people need ways to work together, including operating devices, coordinating agents and clarifying responsibility at work.

He recalls disagreements in strategy discussions, including paused projects that regained support after further validation. He says disagreement can improve team judgment.

He uses “energy” to describe teamwork: disagreements can support exploration when people communicate and support one another, but similar arguments can drain an organization when trust is absent.

Turning to personal life and family, he stresses understanding one’s strengths and recognizing growth and strengths in a partner, children and colleagues.

He describes his entrepreneurial motivation as shaping his own future and continuing to grow, recalling how it extended from personal projects to an automotive company.

He connects wisdom with direct experience of the world, nature and other people, arguing that relationships require time and participation rather than abstract judgment alone.

He hopes agents can handle some repetitive work so employees have more time with customers; any efficiency figures are goals or estimates, not verified outcomes.

He argues that human choices and institutions influence what purposes AI serves, so capabilities and their uses should be considered together.

On close relationships, he emphasizes mutual understanding and boundaries. His account is personal, and people’s needs in relationships can differ.

He says strengths and weaknesses are often connected and that understanding oneself and others requires accepting complexity rather than seeking a single ideal trait.