The Zhitong Finance App learned that Western Securities released a research report saying that as basic model capabilities continue to improve, the competitive focus of agents is expected to gradually shift from the effectiveness of a single question-and-answer session to the stability of task execution, efficiency of cross-application collaboration, and long-term service capabilities. The bank believes that platforms with accumulated advantages in user entry, application ecology, and work context are expected to establish stronger user stickiness in agent competition. Domestic suggestions focus on Tencent Holdings (00700) and Alibaba-W (09988). The two companies each have ecological advantages such as social networking, office, e-commerce, and cloud computing, and are expected to drive agents to high-frequency application scenarios.
The main views of Western Securities are as follows:
Agents are being upgraded from information question-and-answer to continuous task execution, and AI applications are expected to usher in a new round of development opportunities
Next-generation agents, represented by Meta Muse, can complete tasks such as shopping, travel planning, schedule management, and small business management with user authorization using an independent cloud operating environment, long-term memory, and tool connections, and continue to follow up according to changes in external information. Compared with traditional AI assistants, the core change is a shift from providing answers once to taking on tasks over a long period of time, reducing the cost of repeated communication and manual operation for users.
Judging from the product form, major manufacturers are exploring different agent development paths
Muse organizes life and work affairs around users' long-term goals. Grok Bot realizes team division of labor through multiple digital characters with independent responsibilities, and ChatGPT dot coordinates complex tasks across tools and scenarios with unified assistants. Although the three have differences in how tasks are organized and interacted, they all emphasize long-term memory, tool recall, and ability to execute continuously in the background. The bank believes that as basic model capabilities continue to improve, the competitive focus of agents is expected to gradually shift from a single question-and-answer effect to the stability of task execution, efficiency of cross-application collaboration, and long-term service capabilities.
As agent execution capabilities continue to expand, the product boundary between personal and office scenarios is gradually blurring
Codex, Claude Code, and WorkBuddy have extended from software development to research analysis, data processing, document production, and business process automation, intersecting with the capabilities of products such as Muse, Grok Bot, and Dot. In the future, the same task may be completed by different types of agents. When selecting products, users will pay more attention to whether existing data can be accessed smoothly, whether the task can be executed stably, and whether the results can be easily continued to be used. The bank believes that platforms with accumulated advantages in user entry, application ecology, and work context are expected to establish stronger user stickiness in agent competition.
Risk warning: Model iteration progress falls short of expectations; agent product release pace falls short of expectations; privacy and authorization risks, third-party platform restrictions, function opening and commercialization progress fall short of expectations, etc.