The author conducted an experiment to test the limits of AI-assisted coding, known as vibe-coding, by developing over 30 open-source applications. The projects were completed during weekends and evenings, with many tools built as single HTML files or cross-platform Rust applications.
The development list includes a wide range of utilities, such as a Morse code generator, a vector graphics editor, a KeePass-compatible password manager, and a file manager similar to Total Commander. Other projects include a 5 MB disk space visualization tool and a text editor capable of opening files larger than 1 GB.
The author identifies two types of challenges in AI coding: solvable and unsolvable. Solvable issues include technical skill gaps, such as choosing the right language or model, and managing token usage. These skills are expected to become obsolete as AI models improve and integrate these capabilities natively.
Unsolvable challenges involve identifying genuine user needs and market demand. The author notes that while AI can generate code, it cannot algorithmically determine what software is truly valuable or useful to people. Understanding user experience remains a complex, non-algorithmic task.
Monetization presents a significant hurdle, as AI-generated software is easily cloned and redistributed for free. The author suggests that the ease of creating software has led to a devaluation of products, making it difficult to find a sustainable business model in an oversaturated information environment.
