In a recent exclusive interview, Kai-Fu Lee, founder and CEO of 01.AI, discussed the insights from his new book, which explores the AI-driven transformation of business, organizational structures, and the symbiotic future of humans and machines. The conversation focused on the core challenges hindering successful corporate AI adoption, the root causes of frequent failures, and the methodology behind building an "AI Decision-Making Center."
Lee argues that the value of standard software will rapidly decline while customized software becomes increasingly valuable. He introduces a new organizational paradigm for the AI era centered on a Directly Responsible Individual (DRI) mechanism, which redefines the logic of production relations to match AI productivity. This involves restructuring organizations through flattening hierarchies, redefining roles, and overhauling decision-making systems.
According to Lee, the top 10 AI models in China are all of decent quality, making them largely interchangeable for most applications. He identified three main anxieties CEOs face when driving AI transformation, suggesting that a "CEO AI" tool could alleviate these pressures. He emphasized that enterprise AI transformation is a "Number One Project" that cannot be delegated solely to a CIO; treating AI as just another software purchase will only yield limited progress, preventing companies from reaching its full strategic potential. Looking ahead, Lee predicts the fundamental unit of a future enterprise will no longer be a purely human team, but rather a "DRI + AI Agent cluster," empowering a single individual with the effectiveness of an entire professional team.
The core change is AI's enhanced coding ability
When asked about the most fundamental change in AI's landscape this year compared to previous years, Lee stated that the biggest shift is AI's vastly improved coding capability, which directly relates to execution. He distinguished between a Large Language Model (LLM) and an AI Agent by explaining that an LLM "talks and gives advice," while an Agent "takes action," and the core of taking action is running code. He noted that virtually all work in an enterprise, from writing documents and managing accounts to running approval systems and auditing financial reports, essentially involves running code, even if the user doesn't see it.
Lee argued that past debates about AI's value, such as improving programmer efficiency or replacing programmers, were superficial and unimportant. The crucial point is that when AI can write and run its own code, it can directly deliver business outcomes rather than just providing recommendations. For example, an Agent tasked with preparing a board presentation can independently compile company-wide data, anticipate challenging questions, generate responses, and create a complete slide deck, with the entire process executed through code. While we cannot yet simply tell an AI to "double next year's profits," Lee believes this is the inevitable direction. AI will evolve from a "consulting advisor" into an "executor capable of delivering tangible results," ultimately driving business outcomes directly.
The value of standard software will decline, and customized software will rise
Discussing the impact of this execution capability on the software industry, Lee argued that the value of standard software will plummet while customized software will become far more valuable. He explained that companies historically used standard software like Excel or Oracle not because it perfectly fit their needs, but because customization was prohibitively expensive. A custom system perfectly tailored to a specific business required a massive engineering team, a cost only giants like Baidu, Alibaba, or ByteDance could afford. Traditional manufacturing, agriculture, and retail companies had to adapt their business processes to fit rigid software.
Lee believes this is changing. AI can now rapidly understand a company's entire context—its business model, organizational structure, strategic goals, and data assets—and quickly generate a bespoke system. What once required thousands of people can now be accomplished by a small team of a dozen. This means all traditional enterprises can now access the customized digital capabilities previously exclusive to internet giants, freeing them from the constraints of standard software. This also enables 01.AI to provide customized AI solutions without massive human investment. As scalable customization becomes possible, 01.AI's profitability is poised to be on an unprecedented scale compared to enterprise software from the AI 1.0 era. Lee cited real-world examples, including creating tens of millions of dollars in value for a mining company and a partnership with the global agricultural giant CP Group. He anticipates that these lighthouse customer partnerships could generate hundreds of millions of dollars in incremental revenue for 01.AI in the long run—not from cost savings, but from creating entirely new revenue streams that were previously impossible.
A "CEO AI" can reduce anxiety about AI transformation
When asked about the primary concerns of CEOs undergoing AI transformation, Lee shared that based on his interactions with hundreds of leaders from excellent traditional companies worldwide, they face three main anxieties. First, they are uncertain about what to do and how to proceed. Second, they fear making hundreds of AI choices that only produce demos with no real value. Third, they are anxious that AI-native companies or competitors will gain a significant advantage, leaving them at a sudden and severe disadvantage.
To address these anxieties, Lee outlined three things CEOs need to do for a successful AI transformation. First, stop obsessing over which model to use. The question "Which model should I use?" is fundamentally wrong. The top 10 models in China are all good enough and largely interchangeable. Just as having an early Intel chip was useless without an operating system, having a powerful AI model is useless without a thick layer of software that can understand the specific business. Second, create an "Enterprise Map" for AI, which he calls an Ontology. The model doesn't understand the company, and dumping a massive amount of unorganized data into it will cause hallucinations. The business's architecture, processes, and decision-making logic must be clearly mapped to guide the model. Third, wake up the company's most dormant data: meetings. Lee believes a company's most critical data is not financials or sales figures, but "people data," which is largely captured in meetings where strategies are set, decisions are made, and problems are debated. If this meeting data isn't captured, leadership cannot understand what people are thinking, if orders are being executed, or if they are being told the truth. Once this data is activated, along with customer complaints and sales communications, it can be understood through the Ontology framework. The resulting "key" is then handed to the CEO, allowing them to ask any question and get an answer. This is the core concept behind the "CEO AI."
Don't let a CIO's mindset limit AI's strategic value
Lee defended his view that enterprise AI transformation cannot be left solely to the CIO. He acknowledged the CIO's crucial role in ensuring system reliability, security, and stability. However, he argued that a CIO's natural mindset is "tool-oriented," focused on selecting the right software and minimizing risk. If AI transformation is fully delegated to a CIO, the company will treat AI as just another new piece of software, applying it only to low-risk, low-value scenarios like customer service chatbots, legal document review, or meeting summaries. These are minor optimizations that fail to unlock AI's true revolutionary value.
Lee insisted that true AI transformation must be a "Number One Project" driven by the CEO. The CEO must think strategically about how to use AI to restructure the core business logic, boost revenue, improve profit margins, shorten product time-to-market, and reduce key operational risks. Without the CEO's personal involvement, enterprise AI applications will remain stuck at the level of efficiency tools, unable to reach their full potential. He advised that the right entry point is to work backward from core business objectives, not forward from tool selection. Companies should ask, "What is my most important business goal this year, and can AI help me achieve it?" rather than "What AI model or agent system should I buy?" The fastest and most effective transformations, Lee noted, happen when the CEO takes the lead and places AI directly into core operational functions like sales, supply chain, and product innovation. AI is not an IT office tool; it is a core variable for the entire management team to reconstruct business strategy, execution, and organizational structure.
A DRI plus Agent cluster will become the new basic unit of the enterprise
Lee confirmed that his proposed "DRI + AI Agent cluster" paradigm will fundamentally change organizational structure at its most basic level. He explained that the old model relied on headcount growth and hierarchical management. Scaling a business required hiring more people, leading to management inefficiencies and diminishing marginal returns, creating a natural ceiling for human-centric organizations.
In the future, the basic unit of an enterprise will be a "Directly Responsible Individual (DRI) plus an AI Agent cluster." A single core leader with good judgment and strategic direction, supported by thousands of AI Agents, can run an entire business line. Humans will define goals, allocate resources, make key judgments, and handle crucial communication. Repetitive work like execution, coordination, and process management will be fully handled by the agents. Lee sees this not as simple downsizing but as freeing people from tedious management and mechanical work to focus on high-value activities like strategic thinking, creative innovation, business negotiations, and ecosystem partnerships. The result is that small teams can accomplish what previously required large companies, completely opening up an enterprise's capability boundaries. This will also create new roles and growth opportunities, shifting human value from being consumed by process to being amplified by creativity and decision-making.
For small and medium-sized teams looking to implement this capability, Lee introduced the Cube01, a product designed to deliver the organizational intelligence of large enterprises to teams of 10-30 people and super-individuals. The Cube01 is a local, privatized intelligent computing node that integrates 01.AI's multi-agent platform. Its core value is twofold: it localizes data and intelligence, keeping core business data and organizational knowledge on-premises for security and cost control, while using cloud models for complex tasks on demand; and it functions like a "silicon-based employee" that can be integrated into existing office tools like Feishu or WeCom via QR code, adapting to the team's work habits over time. Lee clarified that 01.AI does not manufacture hardware; its contribution is the multi-agent platform's AI capability, while partners supply the hardware.