CICC: AI + code is a product that leads the practical, commercial AI application scenarios.

CICC released a research report stating that recently, the AI+ code application Cursor has attracted market attention with its powerful features. On September 12th, OpenAI released the OpenAI-o1 model, which significantly improved the model's mathematical and programming abilities. In the past two years, the capabilities of AI+ code applications have improved significantly, and many applications have emerged, with Cursor being a representative example. A research report by CSDN in 2023 indicated that nearly 90% of developers in the sample have started using code generation tools, with the vast majority believing that it has improved development efficiency. The team believes that the model's capabilities can enhance the upper limit of AI applications, but understanding the context determines whether the upper limit of AI applications can be realized. AI+ code is a practical product and leading commercial AI application scenario. In major generation tasks, CICC observed that AI+ code is a scenario with a broader and deeper application breadth and depth in terms of practicality, usability, user experience feedback, and penetration rate. CSDN's research report in 2023 indicated that nearly 90% of developers in the sample have started using code generation tools, with the vast majority believing that it has improved development efficiency. For example, GitHub Copilot had over 1.8 million paying users by April 2024, with revenue reaching hundreds of millions of dollars in the 2024 fiscal year. CICC believes that code generation tasks are essentially a type of "standardized" "language" generation work with lower complexity compared to pure text generation, making them suitable for large models to leverage. Cursor is essentially an IDE powered by large models that significantly increases developer efficiency. In terms of product form, most AI+ code products (such as GitHub Copilot, Codeium, iFlyCode, aiXcoder, etc.) are used in integrated development platforms (IDEs) such as VSCode in the form of plugins, while Cursor itself is an IDE that integrates the capabilities of large models, allowing flexible calling of models like GPT and Claude in a locally deployable form. In terms of functionality, CICC categorizes the core functions of AI+ code as "generation" and "interpretation." Besides basic code completion, Cursor also supports line-level, file-level, and project-level code generation, editing, and interpretation. Although there is no significant change in product form, its deep understanding of the development process has led to successful commercial implementation. What inspiration does Cursor's success provide for AI applications? 1) Deep understanding of application scenarios and accurate grasp of customer needs. CICC believes that the improvement of AI application capabilities depends on both the enhancement of the model's upper limit capabilities and a profound understanding of application scenarios and customer needs. It can be said that the model's capabilities can enhance the upper limit of AI applications, but understanding the context determines whether the upper limit of AI applications can be realized; 2) Deep integration with native applications to achieve "invisible applications" of AI. Cursor naturally integrates AI capabilities into applications, maximizing the retention of the original product form. Similarly, in the AI+ text field, Beijing Kingsoft Office Software, Inc.'s WPS AI AI co-writing has significantly improved the experience under the original WPS product form. Risks: - Technological progress falls short of expectations - Industry competition intensifies - Commercialization implementation falls short of expectations.