FROM
PEOPLE
WHO
BUILD.

Give editors a free-text field for icons and you end up with "star", "Star", "star-01" and "⭐" in the same table. So we built a picker instead. Then a design revision asked for 01, 02, 03 next to the steps, and we had to decide whether to fork the plugin or teach it a second trick without turning it into client-specific code.

When I introduced the meta-repo to the team, the goal was a true agentic coding setup: a shared doc set, shared skills, shared rules, all loaded automatically no matter which platform repo you had open. Onboarding got easier and setup went from three commands to one as a side effect of that. Several months later, that's still true — structuring a codebase for AI context turns out to be a different problem than structuring it for human developers. This isn't a pattern we roll out on every project. It's what this specific project needed, given a constraint most codebases don't have: three mature repos that couldn't be merged into one.

If you have engineers on the fence about AI, don't put them on a real project to convince them. Give them their own terminal and let Claude rebuild it. The blast radius is zero and the skill transfer is real.

It's not just a name change. When you call something a performance review, you're already framing it around behavior, effort, and presence. Did you show up? Did you work hard? Are you a good culture fit? The conversation drifts toward personality before it ever reaches outcomes. When you call it an impact review, the question is different from the start: what changed because you were here?

Taking a mobile app apart used to be a craft. It took a specialist, a quiet week, and a great deal of patience. None of that is true anymore.

For two years the story has been the same: AI writes code faster, one engineer does the work of three. I ran the experiment. The typing got cheap almost overnight. What nobody warns you about is what got expensive instead.