On the evening of October 13, at the "Tsinghua University Engineering Management Innovation Talent Education and Development Forum and Tsinghua MEM Course Open Day" event, Zhou Hongyi, founder of 360 Group, delivered a keynote speech. As a doctoral candidate in innovative leadership engineering at Tsinghua University's Department of Computer Science, he combined his technical background with enterprise practice to point out that artificial intelligence is transitioning from the large language model stage to a new phase of intelligent agents. He emphasized that AI agents are more like interns, assistants, or even virtual employees, requiring an approach that treats them as people. He predicted they will transform individuals into "super individuals" and enterprises into "super organizations," with development prospects ten times greater than software, and proposed an enterprise AI transformation framework providing methodology for agent implementation.

"We can no longer treat intelligent agents as simple software or tools," Zhou Hongyi emphasized. "They are more like interns, assistants, or even virtual employees capable of independent work, and we must treat them with the same attitude we would show toward people."

He believes that while traditional large language models possess powerful knowledge bases and language capabilities, they lack true "intelligence" and cannot engage in slow thinking, trial and error, or adjustment when facing complex tasks. This year's three major breakthroughs in the large language model field - enhanced reasoning capabilities, mature open-source free ecosystems, and drastically reduced reasoning computational costs - have collectively driven the birth of intelligent agents.

"True intelligence is not about answering questions, but solving problems," Zhou Hongyi illustrated. When faced with "how to put an elephant in a refrigerator," humans think step by step, attempt, and correct - this "slow thinking" is the core capability of intelligent agents. Through reinforcement learning and task planning, they can decompose goals into executable paths and complete task loops through trial and error.

More importantly, intelligent agents possess four human-like characteristics: autonomous planning, continuous memory, tool usage, and collaborative work. This enables them to break free from the limitations of question-and-answer interactions, proactively invoke professional tools, save working memory, and collaborate with other intelligent agents when facing complex tasks to deliver complete results. "Having one employee do design, write code, and manage finances would certainly not work well," Zhou Hongyi analogized. "Similarly, an intelligent agent should focus on one role." The key to building efficient agent systems lies in "role-playing" and "organizational management."

Zhou Hongyi predicted that intelligent agents will reshape the competitive landscape. In the future, each person could own a "cyber assistant team" of dozens of agents, working 24/7 to process information, write reports, create videos, and manage social interactions, thereby upgrading individuals to "super employees" and "super individuals." Enterprises will also transform into "super organizations." "Many American startups with only dozens of employees can create the benefits of thousand-person teams - the secret lies in the efficient work of intelligent agents." "Software only provides tools, while intelligent agents can directly complete tasks, replacing both human labor and software, catalyzing an entirely new 'agent economy.'"

Facing the agent transformation, Zhou Hongyi proposed a methodological framework for enterprise AI implementation:

First, transform cognition: treat intelligent agents as "people" and clarify role positioning.

Second, start with small entry points: prioritize positions with clear processes and labor-intensive tasks for pilots, such as contract review and market research.

Third, build "virtual teams": handle complex tasks through multi-agent division of labor and collaboration.

Fourth, expert-led approach: have business-savvy backbone staff design processes while AI teams provide technical support.

Fifth, maintain "human in the loop": retain human supervision for critical decisions to ensure controllability and manageability.

Zhou Hongyi concluded with an appeal: "AI will not eliminate people, but it will eliminate those who cannot use AI." Future competitiveness lies in the ability to "lead" intelligent agent teams. He suggested that universities offer courses such as "natural language programming" to cultivate composite management talent with AI collaboration capabilities.