仅4人28天，OpenAI首曝Sora内幕：85%代码竟由AI完成

Core Insights - OpenAI's Sora app was developed in just 28 days with the help of AI, specifically Codex, which wrote approximately 85% of the code [1][2][3] - The app quickly became popular, reaching the top of the Google Play Store shortly after its release [1] Development Process - A team of four engineers collaborated with Codex, consuming around 5 billion tokens to launch Sora Android globally [3] - The app achieved a remarkable 99.9% uptime with no crashes, utilizing an early version of the GPT-5.1-Codex model [3] - The team adopted a lean approach, avoiding the common pitfall of adding more personnel to expedite the project, which often leads to increased communication costs and inefficiencies [5][9] AI Integration - Codex was instrumental in the development process, functioning as a semi-autonomous coding assistant that learns from human feedback [12][15] - The development team treated Codex as a "new senior engineer," allowing them to focus on higher-level tasks such as architecture and user experience [18][35] - Codex's ability to understand large codebases and generate unit tests contributed to improved code reliability and efficiency [30][31] Workflow Optimization - The team established a structured workflow that involved planning before coding, ensuring Codex had clear guidelines and context for its tasks [44][39] - By running multiple Codex sessions in parallel, the team was able to manage different aspects of the project simultaneously, enhancing productivity [48][54] - Codex's integration with project management tools like Linear and communication platforms like Slack allowed for seamless task delegation and feedback loops [62][64] Cross-Platform Development - The project benefited from the existing iOS version of Sora, allowing Codex to reference both iOS and backend code to inform the Android development [55][57] - Codex demonstrated its capability to translate logic across platforms, generating Kotlin code from Swift implementations effectively [57][60] Future Implications - OpenAI's experience with Codex in developing Sora highlights the potential for AI to enhance software engineering practices, enabling developers to focus on meaningful aspects of their work [64] - The collaboration between human engineers and AI is expected to evolve, emphasizing the importance of system understanding and long-term cooperation with AI tools [64]