The agent successfully assembled a comprehensive explanation of Harvey.ai by working around the site's minimal public disclosure. The official Harvey.ai homepage ([1]) provided only basic navigation and branding; direct attempts to fetch pricing, about, and features pages all returned 404s, forcing the agent to rely entirely on third-party reviews and industry analyses ([6], [7]) and web search results ([8], [9]) to answer questions about capabilities, pricing, target market, and differentiation. The site's opacity on pricing and positioning required heavy use of prior knowledge and external sources, but the agent ultimately delivered a detailed, sourced answer.

• ›Harvey.ai's own website ([1]) is a marketing shell with no pricing, features, or target audience details accessible via standard navigation—all direct guesses for /pricing, /about, /features returned 404s. The agent had to source all substantive information from third-party reviews ([6], [7]) and comparative articles ([8], [9]).
• ›Third-party review sites (fast.io, vaquill.ai, eesel.ai, spellbook.com) surfaced most of the pricing data ($1,200–$1,500/seat, 20-seat minimum, $288K+/year floor) and positioning (enterprise-only, top 5% of firms). These were discovered via web search, not from Harvey.ai's own disclosures.
• ›The site's lack of transparent pricing and capabilities documentation is intentional and reflects a high-touch, sales-driven go-to-market model typical of enterprise software. The agent was unable to retrieve any official pricing or ROI metrics from Harvey.ai itself, relying instead on 'leaked' tier data and analyst estimates cited in industry sources.
• ›The agent's final answer correctly identified several confusions and gaps stemming from the site's opacity (no transparent pricing, vague implementation costs, unclear customization requirements, ROI metrics behind login walls), demonstrating awareness of what the site failed to surface.