AI+ Towards a New Intelligent Economy

On May 7, a report titled “AI+ Towards a New Intelligent Economy” was published by Xinhua Daily. This year, the government work report mentioned “intelligence” 17 times and “security” 34 times, marking a significant increase from last year. This indicates that “intelligence + security” has become a dual keyword in national strategy. The report introduces the concept of a “new intelligent economy,” calling for the promotion of a new generation of intelligent terminals and agents, clearly integrating AI into the national strategic layout.

This top-level design signifies that artificial intelligence is transitioning from a stage of technological breakthroughs to large-scale applications, evolving from isolated capabilities to systematic capabilities, and moving from laboratory innovations to industrial infrastructure construction.

The recent popularity of the “Lobster” intelligent agent reflects the latest technological trends, revealing profound industrial logic and user insights. It breaks away from traditional paths and reconstructs the connection between AI and users. Firstly, it eliminates the dependency on “cloud + browser” and opts for instant messaging (IM) as a natural host. Data shows that over 80% of users interact with “Lobster” directly through IM tools without needing to open a webpage, transforming AI from a “tool” into a “conversational partner,” significantly lowering the usage threshold. Secondly, it provides users with a “development experience” through three modular designs: soul (persona), skill (capability), and memory (interaction data), allowing users to train and shape their own intelligent agents, fostering a strong sense of ownership and addressing the pain points of traditional agents that are “fixed in logic and unable to grow.”

From an industrial perspective, its breakout offers three disruptive insights for AI development: for users, it materializes and personalizes the abstract concept of an “intelligent agent,” enabling the general public to intuitively perceive AI’s value, thus promoting the implementation of “AI+”; for the AI industry, it overturns the design logic of “tool-type agents,” proving that the core value of intelligent agents lies in being “growing digital employees” rather than “more complex software”; for the entire industry, it indicates that what software buyers will purchase in the future will no longer be just code, but “intelligent services that solve problems.”

From a technological evolution perspective, it validates the dual-line evolution of AI. The ultimate realization of artificial intelligence does not solely rely on the iteration of large model parameters but involves a parallel progression of “large model evolution + intelligent agent evolution”: large models provide foundational intelligence support, while intelligent agents leverage “tool Scaling Law” (using internet tools, open-source community tools, or even creating new tools through instant programming to promote exponential growth in capabilities), “collaborative Scaling Law” (allowing various collaboration modes during work, such as specialized division of labor, shared responsibilities, supervision, brainstorming, linking, and parallel operations, with millions of intelligent agents on the internet exchanging experiences to generate collective wisdom), and “knowledge Scaling Law” (transforming static knowledge like RAG and knowledge bases into more advanced dynamic skills, realizing the skillization of knowledge), thus converting intelligence into productivity and pointing the way for the realization of Artificial General Intelligence (AGI).

Recent practices show that the large-scale application of intelligent agents often follows a common rule: only when intelligent agents can lower deployment thresholds, form modular capabilities, and support continuous training and collaborative invocation can they truly integrate into business processes. In scenarios such as office collaboration, security operations, and software development assistance, intelligent agents have begun to take on some repetitive tasks, becoming “silicon-based employees.”

The construction of a security system has become the foundational guarantee for the large-scale application of intelligent agents. When artificial intelligence is limited to the dialogue stage, its impact mainly resides at the information level; however, once intelligent agents possess execution capabilities, their influence extends to production systems, organizational systems, and even social systems. The current security challenges faced by intelligent agents are no longer just traditional network attack issues but involve a new risk system formed by the intertwining of model security, data security, and system security. For instance, the “hallucination” problem leads to severe discrepancies in perception and judgment, resulting in destructive operations like erroneous deletions and modifications; excessive system permissions and unpredictability make it impossible to assess whether subsequent actions pose dangers; most critically, intelligent agents lack the ability to identify harmful skills, making them susceptible to malware disguised as normal skills during the autonomous download and learning of new skills, which undoubtedly plants a time bomb for our system security, posing significant potential hazards.

The industry is exploring a security path of “using models to govern models,” which involves utilizing security intelligent agents to identify attacking agents, analyze abnormal behaviors, and automatically repair system vulnerabilities. By constructing an integrated intelligent security capability that covers models, data, and execution environments, replicable security solutions for intelligent agent applications can be provided. The key to this security system lies in achieving dynamic protection through “intelligent agents countering intelligent agents,” adapting to the security needs of the new intelligent economy.

Innovating inclusive pathways is crucial for promoting the large-scale application of intelligent agents. Currently, three main factors restrict the large-scale application of intelligent agents: high usage thresholds, high deployment costs, and insufficient security trust. To address these issues, the industry is exploring a layered promotion approach that provides “out-of-the-box” intelligent agent capabilities for ordinary users, opens capability interfaces for developers, and offers customizable industry solutions for enterprises, enabling different groups to find suitable application methods for their scenarios. For example, 360, leveraging its twenty years of accumulated security expertise, has explored an innovative path of “security + AI” by constructing multiple types of security intelligent agents to accurately identify and alert against new risks such as injection attacks and permission breaches, achieving automatic blocking of attacking agents and system repairs.

In the AI era, competition fundamentally revolves around ecosystems, scenarios, and inclusivity. As the number of intelligent agents reaches the tens of billions, artificial intelligence is transitioning from a technological variable to a production variable, evolving from an innovation factor to a foundational infrastructure. Future competition between enterprises may no longer be just about human resources but about the collaborative capabilities of “carbon-based employees + silicon-based employees”; competition between industries will also shift from a singular technological capability to competition within the intelligent agent ecosystem.

Looking ahead, promoting the development of a new intelligent economy requires focusing on three directions: first, deepening the transformation and upgrading of service industries; in the past, internet software and hardware were designed for humans, while in the future, intelligent agents will become the main force of the internet, with e-commerce, browsers, databases, and search engines fully adapting to the operational logic of intelligent agents, forming a new digital foundational software ecosystem; second, enhancing the systematic reconstruction of industry security; facing the machine-level confrontations brought by “hacker intelligent agents,” it is essential to deepen the concept of “using models to govern models” and strengthen the construction of the security intelligent agent ecosystem, continuously improving the dynamic protection system; third, increasing the integration of intelligent agents with emerging industries, enhancing the practical applications of intelligent agents in fields such as embodied intelligence, scientific research, drones, and connected vehicles, creating human-machine collaborative demonstrations to improve organizational efficiency and innovation capabilities.

In the context of the continuous advancement of “AI+”, a new intelligent economic form centered around intelligent agents is accelerating its formation. This is not only an inevitable trend of technological development but also an important opportunity for promoting high-quality development. Through a balanced approach of technological innovation and ecosystem construction, along with coordinated development and security, artificial intelligence will truly become a universal capability infrastructure, akin to electricity and the internet, injecting strong momentum into the high-quality development of the digital economy.