Recently, there has been a lot of discussion around a simple question:
Should AI artifacts default to Markdown or HTML?
Markdown is simple, readable, and token-friendly.
HTML is heavier, but it can represent information in a much richer way. With HTML, an artifact can have proper sections, tables, cards, callouts, layouts, visual hierarchy, and even interactive components. While HTML may consume more tokens upfront, the additional structure and clarity can reduce ambiguity and back-and-forth later.
My thinking went in a slightly different direction:
