In the very first days of Gemini's design, I was arguing with other implementers for using a large subset of CommonMark (no raw HTML, no inline images, but pretty much everything else) as the preferred content-type to serve over Gemini. In the end, the goal of ease of implementation for client authors won out. The main arguments were that while there are Markdown parsers for many languages, most of them are narrowly aimed at converting to HTML; Markdown has known ambiguities; the line-oriented markup we ended up with was a strict improvement over Gophermaps while remaining equally simple to parse.

I also miss inline emphasis and other kinds of text-oriented typography, but the reasons gemtext omits them are fairly good.