Vibe Coding as a Coding Veteran. From 8-Bit Assembly to English-as-Code
13 days ago
- #Tower-of-Hanoi
- #AI-assisted-coding
- #programming-productivity
- The author explores 'vibe coding,' where AI assistants handle coding tasks, allowing developers to focus on design.
- With 40 years of coding experience, the author tests AI coding assistants (OpenAI’s o3, Anthropic’s Claude Sonnet 4, Google’s Gemini Pro 2.5) over 40 hours.
- The project involves a Python-based Tower of Hanoi solver (~5k LOC) implementing various AI search algorithms.
- AI assistants demonstrated impressive understanding of code and natural language, often surpassing human speed and accuracy.
- Claude Sonnet 4 emerged as the preferred assistant for complex coding tasks due to deeper engagement and insight.
- Despite AI strengths, flaws included overly complex refactoring, incorrect assumptions, and subtle bugs requiring human oversight.
- Productivity gains were significant (2X speed), but AI-generated code required careful review to ensure correctness.
- The author reflects on the psychological impact of AI-assisted coding, noting both empowerment and existential questions.
- Natural language as a programming tool introduces ambiguity, raising concerns about precision in mission-critical applications.
- The experiment concludes that AI coding assistants are powerful but require skilled developers to guide and validate their output.