Meta has developed an artificial intelligence system to enhance the efficiency of working with large data streams, which has significantly improved code navigation and reduced the number of tool calls.
Meta has developed an artificial intelligence system to enhance the efficiency of working with large streams of data, significantly improving code navigation and reducing the number of tool calls.
Meta introduced a system that uses artificial intelligence to handle vast data streams encompassing over 4,100 files in three programming languages. Initially, the AI agents exhibited insufficient speed and quality in editing complex code. In response, a pre-computation system was developed with more than 50 specialists.
This system increased code coverage for the AI from 5% to 100%, providing structured navigation guides for all modules. Context files contain between 25 to 35 lines of information, including command descriptions and key files, enabling effective handling of development and data management tasks.
Amidst growing competition in the AI field, Meta stands out with a unique methodology for working with code, which has become particularly important in the context of rapid technological advancement and the need for compatibility among different systems. The company emphasizes the significance of integrating knowledge into the workflow, which can substantially enhance productivity.
The system not only optimized development but also improved code quality. Interim tests showed a 40% reduction in tool calls and improved quality according to independent expert assessments. Tasks that previously took two days are now completed in just 30 minutes, opening up opportunities for scaling such solutions in other companies.
However, the results raise questions about the role of contextual information for AI. Previous studies claimed that context files could reduce the agent's efficiency, which requires further investigation. In the case of Meta, the proper organization of context files significantly enhances the AI's interaction with code.
Thus, the implementation of AI in code work demonstrates the company's systematic approach and the importance of tribal knowledge. In the future, similar methodologies may become standard in the industry, changing approaches to software development.
Sources
Replies (0)
No replies in this topic yet.
