In an interview with IBM Think, Goodhart shared his unique approach to leveraging AI for coding, which varies based on project complexity. For greenfield projects, he embraces full automation, starting with a detailed textual description of the project. This serves as a foundation for collaborative ideation with the AI, allowing it considerable autonomy during development. He later refines the initial unedited code through targeted adjustments. For small feature enhancements or debugging, he adopts a more controlled method, providing concise descriptions to the AI and closely monitoring the output to ensure issues are accurately reproduced. Complex features require a balanced approach, where he invests more effort upfront and collaborates iteratively with the AI, maintaining tighter control. Overall, Goodhart’s strategy illustrates that the ambiguity of the problem dictates the level of AI autonomy, enhancing productivity while mitigating risks in critical code integrations.
Source link
