Yale Scholar Zhou Hongyi Reveals Continuous AI Programming for Over Ten Hours

On March 24, a delegation of over a hundred renowned entrepreneurs from the Yale Scholars Alumni Group visited 360 Group to gain an in-depth understanding of the company's technological layout and innovative achievements in the fields of cybersecurity and artificial intelligence. During the event, Zhou Hongyi, the founder of 360 Group, shared insights as a Yale alumnus, discussing the technological evolution and enterprise implementation in the era of OpenClaw ("Lobster").

Zhou Hongyi pointed out that artificial intelligence is transitioning from the "Year of Big Models" and the "Year of Intelligent Agents" to the "Year of Lobster" driven by OpenClaw. Unlike previous large models that focused on dialogue capabilities, intelligent agents possess execution capabilities, enabling them to call tools, operate software, and complete complex tasks, marking a shift in artificial intelligence from "being able to speak" to "being able to do."

He introduced that "Lobster" represents a more advanced form of intelligent agents that can run locally and obtain higher system permissions, with capabilities for tool invocation, autonomous learning, and continuous memory, while also exhibiting distinct personalized characteristics that allow them to continuously understand user needs and optimize performance during use. This gives them strong application potential in scenarios such as content production, development assistance, data analysis, and cross-system process automation.

In this context, the form of software and development models are also changing. Zhou Hongyi believes that traditional software will shift from being "pre-manufactured products" aimed at humans to atomic "skill modules" designed for intelligent agent invocation, which can be flexibly combined based on task requirements; when existing tools cannot meet demands, intelligent agents can even generate code on their own to complete tasks. Thus, the software production method is transitioning from "pre-developed" to "on-demand generation," which will have a profound impact on the software industry structure.

He further pointed out that as the technical threshold lowers, the role of software engineers will change from merely writing code to scheduling, orchestrating, and designing capabilities for intelligent agents, and business managers will also need to build a new capability system that adapts to the intelligent era.

Regarding the implementation issues for enterprises, Zhou Hongyi emphasized that the current "Lobster" still possesses certain randomness and uncertainty; directly applying it to core business processes that emphasize certainty may pose risks. Therefore, enterprises should adopt a gradual approach. To this end, he proposed "three major infrastructures" for the application of Lobster in enterprises: on one hand, customizing skills and orchestrating processes based on business scenarios to transform general intelligent agents into compliant "professional intelligent agents"; on the other hand, building a unified management platform to achieve permission control, task scheduling, and multi-agent collaboration. Additionally, it is necessary to consolidate enterprise knowledge, industry standards, and SOP into a callable "enterprise memory" to enhance the specialization level of intelligent agents.

![](https://cdn.pingwest.com/portal/2026/03/24/portal/2026/03/24/Mp0Nfe4YZT610BbBxYKF_55R723nd8N7? In terms of security, Zhou Hongyi pointed out that intelligent agents, due to their high permissions, face risks such as prompt attacks, malicious skill poisoning, and data leaks. In response, 360 proposed the security concept of "modeling with models," which involves real-time monitoring and semantic analysis of intelligent agent behavior through a large security model, releasing its application value while ensuring a safety baseline. He suggested that enterprises could initially deploy intelligent agents through cloud services or dedicated devices, and strengthen monitoring and auditing capabilities to avoid risk spillover.

At the event, Zhou Hongyi also shared his personal experiences: "I have been programming with AI almost every day recently, sometimes even for more than ten hours straight, without time to rest or check WeChat, continuously giving instructions to AI via my phone to develop various intelligent agents." He stated that many business leaders in Silicon Valley are also personally using AI tools and even coordinating multiple intelligent agents to work together simultaneously.

"These business managers do not lack programmers; the reason they are personally getting involved is that everyone realizes that without deeply using AI themselves, they cannot truly understand where this round of technology will lead the industry." Zhou Hongyi said, "Without firsthand experience, many strategic decisions at critical junctures cannot be made."