The AI Asia Vets product is forming an evolution path of infrastructure and application synergy characterized by "upgrading underlying model capabilities + flourishing intermediate tools + landing commercial scenarios."
According to reports from Zhitong Finance APP, Southwest Securities has released a Research Report stating that currently, the level of AI development is evolving from reasoning agents to intelligent agents. AI products are gradually able to understand objectives, have external memory and reasoning capabilities, and the related Industry Chain for intelligent agents is undergoing a systematic leap from model capability enhancement to application commercialization. The AI intelligent agent products are forming an evolutionary path of "upgrading underlying model capabilities + flourishing intermediate tools + landing commercial scenarios" in the collaboration of infrastructure and applications. In the future, AI intelligent agent applications will need to further enhance planning capabilities, possess better memory, and have stronger multimodal understanding capabilities to unlock monetization potential.
The main points from Southwest Securities are as follows:
The stages of AI development are shifting from reasoning agents to intelligent agents, with an enhancement in the intelligence level of the model foundation.
Currently, the level of AI development is evolving from reasoning agents to intelligent agents. AI products are gradually able to understand objectives, have external memory and reasoning capabilities, and the related Industry Chain for intelligent agents is undergoing a systematic leap from model capability enhancement to application commercialization. The capability of large AI models is advancing through three expanding curves during pre-training, post-training, and testing phases, where pre-training establishes the internal intelligence limits of the model, while post-training and testing phases help unlock the model's potential in specific fields and reasoning aspects. Currently, the iteration of the foundational model is slowing down, gradually shifting from training expansion to testing expansion, leading to a transition in the primary and secondary curves. This results in a decreased dependency on large-scale clusters, an increasing demand for reasoning computing power, and a greater focus on the commercialization capabilities of AI products and ecological development.
Intermediate tools are accumulating insights, and the developer ecosystem is actively being constructed.
In the intermediate layer, the communication protocols and development tools required for the intelligent agent ecosystem are rapidly emerging. Representative technologies such as the Anthropic MCP protocol and Google's A2A protocol are assisting in building a new operating system for intelligent agents, establishing a unified interaction interface between models and tools, as well as between intelligent agents. Among these, the number of server creations on the MCP Server discovery platform Smithery in March 2025 saw a threefold increase compared to February, and A2A has gained support from over 50 partners, accelerating the prosperity of the developer ecosystem. The standardization of development tools and underlying frameworks can be likened to the USB-C interface for mobile phones in the Internet age, or compared to the Android API used for communication between apps and operating systems, which will accelerate the commercialization process of AI intelligent agents.
The initial products are generating revenue faster, with the dawn of commercial applications becoming evident.
At the application level, intelligent agent applications are divided into cross-industry general products and vertical professional products. The former is relatively mature, with some products having begun large-scale applications, while the latter started commercialization a bit later but is expected to become an important lever for B-end digital transformation. Currently, intelligent agents are rapidly landing as interactive AI products, with initial products like Cursor and Glean already achieving over a hundred million dollars in annual recurring revenue (ARR), demonstrating high long-term growth potential, and new charging models based on actual delivery results and task completion rates have emerged.
Overall, AI intelligent agent products are forming an evolutionary path of infrastructure and application synergy characterized by "upgrading underlying model capabilities + flourishing intermediate tools + landing commercial scenarios," with future AI intelligent agent applications needing to further enhance planning capabilities, possess better memory, and have stronger multimodal understanding abilities to release monetization potential.
Related symbols.
1) Computational power: NVIDIA (NVDA.US), Broadcom (AVGO.US); 2) Intermediate tools and data layer: Google (GOOGL.US), Snowflake (SNOW.US); 3) Downstream applications: Salesforce (CRM.US), SAP, Shopify (SHOP.US); 4) Cloud Computing Service: Amazon (AMZN.US), Microsoft (MSFT.US), Google.
Risk Warning
Risks including AI technology progress not meeting expectations; AI commercialization progress not meeting expectations; investment returns not meeting expectations.
