The agent successfully gathered comprehensive information about Reka.ai by combining a single usable homepage fetch with extensive web search, since the website itself had broken internal links and no pricing page. The agent pieced together what Reka does, who it's for, how it's priced, and competitive differentiation—but had to work around significant site navigation gaps and rely heavily on third-party pricing aggregators and news sources rather than first-party documentation.

• ›Step [1] (homepage) provided only metadata summary in the page description ('natively multimodal models (Spark, Edge, Flash, Core) for video, image, audio, and text. Used by enterprises in security, media, and defense'); the actual page content was not included in the fetch response, forcing reliance on external sources.
• ›Steps [2–10] all returned 404s (pricing, products, products/vision, products/research, about, blog, docs, api), indicating the website has minimal navigable structure—most standard business pages don't exist or are broken, creating a hostile environment for agent discovery of standard information (pricing, product docs, company background).
• ›Steps [12–15] (web searches) returned third-party pricing aggregators (tokenando.ai, eesel.ai, pricepertoken.com, usagepricing.com), news articles (Decrypt), and an arXiv paper—none from reka.ai's own domain. The agent had to reconstruct the answer from scattered external sources rather than authoritative first-party documentation.
• ›The agent explicitly cited 12 sources in its final response; only 1 was from reka.ai's own domain (the homepage), indicating 91% of the answer came from outside Reka's published materials, not from their website navigation.
• ›Site is poorly agent-ready: no pricing page, no dedicated product pages, broken internal links, and limited content depth on the homepage mean an agent cannot self-serve even basic commercial and product information without external research.