Hexmap: Week 0x21

This week was mostly about performance. Embolded by my new comfort with “agent” coding, I used Kilo Code (with many models: Gemini 2.5, Claude Sonnet 4, and GPT-4o) to implement a caching layer for map tiles (the work done for the many requests from the slippy map during rendering), and then added map context (the work done for a single page).

I had to micromanage quite a bit to find and resolve edge case problems, and spent an hour essentially arguing with an architect agent about approaches. The end result is something I’m happy with but nonetheless feel like another developer “owns”. It’s very much like having a developer on a team who gets the work done and passes the tests; even after code review, it still feels like their code and they remain the point person for future questions about it.

Unrelated while debugging some rendering issues, I found myself doing verification against WOF data. While there, I looked up the number of Wikipedia references, versus the total, for the placetypes I’m supporting: