Why Specialized AI Agents Will Outpace General Ones – Insights from Zhou Hongyi

On July 23, 2025, during the opening ceremony of the 24th China Internet Conference, 360 founder Zhou Hongyi shared his views on the development path, security challenges, and application prospects of AI agents, emphasizing that they are moving from concept to practice and that China’s abundant application scenarios create a development opportunity for all industries.

Zhou noted that the debate between “general” and “specialized” agents shows that specialized vertical agents have more vitality; under current large‑model technology, general agents lack sufficient generalization, and even programming agents require continuous business‑driven training.

He likened future agents to “virtual experts” with domain specialization, comparing them to doctors and lawyers, and cited examples such as accounting agents, clerical agents, and contract‑approval agents that quickly solve real problems and lower enterprise adoption barriers.

Zhou highlighted China’s unique advantage of having the world’s richest variety of industrial categories and scenarios, from manufacturing to government services, content creation to cybersecurity, and illustrated this with 360’s Nano AI video‑generation feature that can turn a script into a finished video in minutes.

He warned of three major security risks: hallucination and erroneous output that could cause severe consequences in critical fields; lowered barriers for instruction‑injection attacks allowing ordinary users to manipulate agents; and the emergence of “agent hackers” who can control hundreds of agents to launch attacks.

To address these challenges, 360 built a security‑expert agent and developed a “Model Guardian” that monitors input and output, combines enterprise knowledge bases, and mitigates hallucination risks, enabling real‑time detection and defense.

Regarding the talent shortage in cybersecurity, Zhou argued that agents can serve as virtual security experts, stating that a unit with sufficient compute can deploy 20‑30 security agents, equivalent to dozens of human experts.

He predicted that the proliferation of agents will transform human roles, turning individuals into “super individuals” who define, plan, manage, and supervise digital teams of agents, freeing them from repetitive labor and requiring strong expression and planning abilities.

Zhou outlined 360’s four initiatives: (1) full‑process automated defense in security, creating agent teams to replace human experts; (2) turning browsers into AI browsers with real‑time translation and analysis; (3) transforming search from keyword queries to question‑driven solutions and deliverables; (4) launching China’s first AI‑agent community, hosting over 50,000 agents representing various professional fields, and an upcoming AI‑agent factory that lets non‑programmers create specialized agents using dozens of leading Chinese large models and AI tools.

He declared 2025 the “Year of AI Agents,” a critical period for technology deployment, stating that large models are the engine and agents are the vehicle to solve real problems, and as costs continue to fall, agents will enter more industries, with enterprises advised to start with simple scenarios like accounting to gain early advantage.