Lessons From Rolling Out an AI Harness on a Legacy Codebase
Review comments per PR fell from 5.3 to 0.35. A backend module estimated at 7 to 10 days shipped in 2. The most useful lessons, though, came from the places where I was wrong.
ideas worth sharing, lessons worth remembering
Review comments per PR fell from 5.3 to 0.35. A backend module estimated at 7 to 10 days shipped in 2. The most useful lessons, though, came from the places where I was wrong.
Developers are leaving the IDE. Designers are writing Claude Code skills. Product managers are shipping PRs instead of Jira tickets. I watched all three happen in the past month.
After nearly two decades of building products behind closed doors, I'm opening up. Here's what changed, and why now.
After building and leading teams for over a decade, here are the patterns that separate great teams from the rest. Spoiler: it's not about the tech stack.
Not the hype version. The practical reality of integrating AI tools into a 19-year workflow of shipping products.
I gave 50+ technical talks but never wrote them down. Here's what that taught me about shipping your ideas vs hoarding them.
I interviewed 70 developers in one year. The JavaScript questions were the least interesting part.
What happens when you try to map the technical knowledge of 2,500 people across 150 products. The tools were the easy part.