What is the quality of software that AI writes?

This article examines the quality of software produced by AI coding agents, drawing on the author's experience with GPT 5.5 and 5.6 in Python projects. It acknowledges the productivity boost but highlights issues like code bloat (2-3x more code than necessary), difficulty in simplification, and lack of instinct for structural clarity. The author discusses the debate on whether source code will become obsolete, but argues for human-readable code for debugging, auditing, and reasoning. It also notes that code quality principles are broadly agreed upon and some can be measured programmatically. The article provides practical insights for developers using AI tools, emphasizing the need to actively guide and refine AI-generated code to avoid technical debt.