Dorodango (blog.fsck.com)

A developer shares insights on two distinct approaches to AI-driven software creation, contrasting detailed design processes with iterative, minor adjustments. The first method involves extensive pre-planning and documentation, often culminating in a lengthy implementation plan executed by AI models like Claude or Codex. This "fast waterfall" development emphasizes a comprehensive design phase, allowing the AI to iteratively troubleshoot and refine the project until a successful implementation is achieved. The process culminated in a successful run after multiple attempts, showcasing the potential of AI as a reliable programming partner. In contrast, the second approach, akin to the Japanese art of Dorodango, focuses on making small, impactful changes to existing software. This "polishing" workflow allows for quick adjustments rather than full-scale rebuilds, responding to user requests without overhauling the underlying architecture. The developer finds beauty in this process, comparing it to polishing a rough mud ball into a glossy sphere, while challenging misconceptions that coding agents produce chaotic code. This duality in development styles highlights the evolving role of AI in software engineering, encouraging the community to appreciate both comprehensive and nuanced workflows.