The Teaching Computer (build.ms)

In a recent talk, a developer shared their experience of utilizing an AI tool, Codex, to automate the refactoring of their app, Plinky, while they slept. By creating a list of small tasks in a document titled "Plinky Small Wins," they allowed Codex to generate manageable pull requests overnight. This innovative approach not only streamlined the code review process but also demonstrated how AI excels when given well-defined, narrow tasks. The developer praised Codex for producing quality refactorings and even suggested improvements, emphasizing the importance of maintaining human oversight in the workflow. Additionally, the developer employed Codex's Computer Use plugin to automate testing of deeplinks within the app. Although the AI faced some challenges with the iOS simulator, it creatively generated a simple HTML page to facilitate testing, showcasing its ability to think outside the box. The experience underscored a crucial insight: providing AI tools with clear context and constraints leads to superior results. The talk served as a reminder of the evolving role of AI in software development, emphasizing that collaboration with AI is about defining the solution space rather than detailed task instruction, ultimately enhancing productivity and innovation in the coding process.