I started to think differently about web search for agents, instead of traditional index, I would prefer to have an agentic WALKER that literally walks through pages and finds out whats needed, even if seeded with some initial web search.
I needed to find a word on a wayback archived website. Could not. So I scraped the waybackmachine pages to .md… and found it that way… sorry Archive.org
I like the idea, but its a surefire way of bloating the context. Although I guess you'd want a subagent to do the search.
Wouldn’t searching an index be faster? Unless you’re really digging through a site for some super deeply nested link?