AI-driven coding assistants are transforming software development, yet their full potential in Integrated Development Environments (IDEs) remains underutilized. A key challenge is their tendency to hallucinate, producing plausible but incorrect code and leading developers down unintended paths. Current static file-based IDEs also lack support for tracking the provenance of AI-generated code or integrating version control in ways that match the dynamic and iterative nature of AI-assisted workflows. Consequently, developers lack tools to systematically manage, refine, and validate Generative AI (GenAI) code, making correctness, maintainability, and trust difficult to ensure. Inspired by the art of Japanese Bonsai gardening—emphasizing balance, structure, and pruning—we propose a new paradigm: an IDE where AI is free to generate, and developers guide evolution by pruning and shaping alternatives. We present BonsAIDE, a prototype that supports branching, comparison, and pruning of AI-generated code. In an initial study with ten participants, we observed: (1) diverse exploration strategies across identical tasks; (2) high tool acceptance with low perceived difficulty; (3) benefits of branching and pruning, including clutter reduction and parallel exploration; and (4) concrete feedback on desired features such as side-by-side diffs and improved navigation. These findings motivate future research on provenance, prompt-aware navigation, and scalable human–AI interaction.
Moreno-Lumbreras et al. (Tue,) studied this question.