Articles about software engineering, architecture, and professional software development.
Dex Horthy says agent-built codebases start to struggle after three to six months, and he admits nobody can prove it either way. I went looking for that decay in three of my own repos, one of them almost 100% agent-written. I didn't find it, and I think the reason is the process around the agents. This is my experience with my own repos, no rebuttal intended.
Documentation-heavy processes like RUP were good engineering that died on human nature. AI agents are obedient and meticulous, so I rebuilt OpenUP, RUP's lean survivor, as a framework that makes Claude Code work one logged, traceable task at a time.
A safe AI chat companion for kids, 545 commits, 97%+ test coverage, and only 1.7% human-written code. The overview of the Kaze project and what it takes to build production software with AI agents.
Cloning a React toast library into a Rails-native gem with Stimulus. What Codex handled, what still needed engineering judgment, and why Stimulus was the right fit.
How code review changes when AI generates most of the code. What to focus on, what stays the same, and why professional judgment matters more than ever.
The story of how I built this website without writing code, using AI Vibe Coding through Cursor. Reflection on how AI helped me overcome the blank page problem and build something real.