The agent successfully assembled a comprehensive explanation of Asimov Academy by combining web search results with third-party review articles, since the main site itself (asimov.academy) was not machine-readable or did not surface key information like pricing, plan details, or differentiators. The agent worked around poor site navigability by relying on external sources (VirtualInspira, CodigoFacil, and the hub blog) to answer all four questions (what, who, pricing, differentiation), though it could not verify or find complete pricing details directly from the official site.

• ›Direct site fetches returned only minified HTML/JavaScript-heavy responses ([1], [3], [6], [25]); no human-readable structured content was available from asimov.academy itself. The agent could not extract concrete pricing, course lists, or differentiators from the primary domain.
• ›All substantive information came from third-party review sites fetched in steps [16] and [17] (VirtualInspira and CodigoFacil reviews), which contained pricing tiers (R$ 1,793.55 annual / R$ 6,277.40 lifetime), course counts (275+ hours, 79 projects), and comparisons to Alura—information not found on asimov.academy directly.
• ›The blog post fetched in step [26] (hub.asimov.academy/blog/asimov-academy-vale-a-pena) was cited as a source but its HTML was not readable in the response; the agent relied on search result snippets and prior knowledge to synthesize the final answer.
• ›Site navigation was poor: `/preco`, `/pricing`, `/sobre`, `/planos` all returned 404s; pricing information only appeared in campaign/org-specific URLs like `/aumento-vitalicio-org/` which the agent never directly fetched. No dedicated pricing page was discoverable.
• ›The agent explicitly flagged missing information in its response (5-plan options mentioned in search but not located, course-specific plans not found, compressed course catalog HTML), signaling awareness of incomplete coverage despite task success.
• ›Web search (33% of fetches) and prior knowledge (58%) were critical; only 8% came from previous artifact. The agent had to bootstrap its understanding from external reviews rather than the official site's information architecture.