news.ycombinator.com• 3d ago• 1 min read
Show HN: Neither HTML nor Markdown is enough: a way out of the AI doc dilemmaA recent post from Anthropic's Claude Code team ("The Unreasonable Effectiveness of HTML", https://thariqs.github.io/html-effectiveness/) argued that with today's context windows the bottleneck is human attention — rich HTML for decision surfaces, Markdown for durable records. That split makes sense to me. What it leaves open, in my experience, is the durable record itself: what happens when that is the thing an agent must iteratively co-author? You end up picking one and paying for the other — a surface people will read, or a record you can keep — and whichever you pick, the other drifts.Markdown treats the document as one long flat string — an agent has to quote context back just to say where an edit goes. The case against HTML I don't need to make: that post's own thread ran 274 comments (https://news.ycombinator.com/item?id=48071940). Mine is narrower — HTML is generated, not maintained. Markdown fails at addressing; HTML fails at being the record.That gap is what GEML is for: a pl