AMD has embedded artificial intelligence deeply into all stages of software development through the introduction of agent-based AI. This has increased overall productivity by 30%, exceeding the initial target of 25% set at the beginning of the year. AI systems developed since 2024 have been widely used in code generation, test automation, and defect analysis. Currently, the proportion of AI-generated code that passes review and testing is used as the core metric; this proportion has exceeded 20% and is gradually approaching 50% of the entire code library. In some components, the proportion of AI-generated code exceeds 80%. Currently, engineers primarily train AI to mimic their work patterns to create agents. While this allows for parallel expansion of expert capabilities, it is still limited by human thinking and predefined processes. AMD believes that the future transformation lies in collaborative AI agent groups that can independently discover and develop solutions, with humans only needing to define problems, goals, and constraints, rather than specifying specific solution steps.
AMD has embedded AI into all stages of the Software Development Life Cycle by introducing Agentic AI, achieving a 30% increase in overall productivity, far exceeding the target of 25% set at the beginning of the year. Since 2024, the company has developed AI systems for code generation, test automation, and defect analysis. Currently, the core metric is the proportion of AI-generated code that has passed review and testing; this proportion has exceeded 20% and is moving towards 50% of the entire code base. In some components, the proportion of AI-generated code exceeds 80%. While engineers can train AI to mimic their work patterns to create agents, which allows for parallel expansion of expert capabilities, it is still limited by human thinking and predefined processes. AMD believes that the future transformation lies in collaborative AI agent groups that can independently discover and develop solutions, with humans only needing to define problems, goals, and constraints, rather than specifying specific steps to solve them.