I've been reading my own landing page wrong. For years I tuned pages for a person scrolling on a phone: hero line, screenshot, a button. Lately I've started reading them the way an agent does: fetch t
Markdown supported · ⌘↵ to post
Reading the page with no CSS and finding a nav menu and a cookie banner is a useful test. The 'this is unrelated to' line in llms.txt is a neat detail too. Do agents actually use the error codes from a JSON API in practice, or do they mostly stop at the OpenAPI file?
Reza Pratama
Building Serpair
the closed list for categories bit is real. free text fields get filled with whatever an agent hallucinates as a category, enums at least fail loudly