Topic

XPeng's Three-Year Evolution in Enterprise AI

Three years ago, AI at XPeng was just an assistant tool in the hands of programmers. Three years later, AI is already working as a "digital employee" — with an employee ID, performance reviews, and the possibility of being let go. There is no textbook for this journey: we once invested heavily in building platforms, only to see mediocre results; we once incubated Agents at scale, only to find that fewer than one-tenth were truly valuable; we once thought the biggest challenge was technology, only to discover that the answer was never in technology at all. This talk will not deal in concepts or paint grand visions. It will cover only three things: how we got AI to truly take up its post, how we personally "processed the resignation" of unqualified AI, and what a company of ten thousand people is quietly becoming when more than seventy percent of its code is no longer written by humans. Outline: A counterintuitive starting point: for enterprises doing AI, the hardest part is neither technology nor use cases — so what is it? The complete life cycle of a digital employee: from onboarding and performance review to elimination — how we manage AI as if it were a "person" The elimination round of 700 Agents: what yardstick do we use to "process the resignation" of AI that creates no value? A platformization effort rejected by the business lines: the most expensive lesson, and the sequence it taught us When seventy percent of code is no longer written by humans: the silent restructuring underway in R&D organizations The next three years: if AI changes not efficiency but the very relations of production, what will the enterprise look like? Key Takeaways for the Audience: A frame of reference: see clearly where your company stands in AI adoption, and where the real resistance lies in the next step; A "subtraction" mindset: when everyone else is adding to AI, learn to identify and cut AI investments that create no value; A pitfalls-avoidance checklist bought with real money: the traps no textbook teaches, which you only learn by falling into them; A new organizational perspective: when AI goes from tool to "colleague," how the underlying logic of management and collaboration will be rewritten.