The agent successfully assembled a comprehensive explanation of Novu's purpose, pricing, target users, and competitive positioning, but had to rely heavily on web search and prior knowledge because the site's main pages (homepage, pricing, comparisons) returned truncated or inaccessible HTML. Of the five pages the agent actually fetched from novu.co, only two returned usable content (the comparison and blog pages); the rest were either raw Next.js bundles or 404s. The agent filled gaps by citing external sources (SoftwareSuggest, APIScout, docs) and applying domain knowledge about notification platforms.

• ›Step [1] (homepage) and [3] (pricing) both returned raw, truncated Next.js HTML with no readable content, forcing the agent to abandon the site's primary information sources.
• ›Steps [4], [6], [7], [10] returned 404s for /why-novu, /about, /features, and /comparison—suggesting either poor site structure or the agent guessing at URL patterns that don't exist.
• ›Step [14] (/comparison/building-in-house) and [17] (/blog/notifications-system) were the only novu.co pages that returned meaningful content, both discovered via web search results rather than site navigation.
• ›Step [20] (docs.novu.co) provided structured documentation but was accessed via search referral, not site navigation.
• ›The agent had to rely on external third-party sources (SoftwareSuggest, APIScout, SourceForge) for pricing details, competitor comparisons, and use-case clarification, indicating Novu's own site does not publish this information in a directly fetchable, machine-readable form.
• ›Web search was essential: 44% of fetches came from the previous artifact (the agent's own reasoning), 22% from prior knowledge, and 22% from web search—suggesting the site itself was not self-contained or crawlable.