horizon-ai.ai
success
Claude Code
intent

What does horizon-ai.ai do and who is it for? Explain it back to me.

12steps
66.1sduration
$0.6760cost
121,258tokens
19 steps7 reasoning steps4 searches
home
docs
docs
/services
search
search
/en
docs
/case-studies
search
/company/horizon-ai-tw
search
100%
on-site discovery
100%
reliability
63%
link following
path origin
  • previous resource63%
  • prior knowledge38%
insight

The agent partially satisfied the task by assembling a comprehensive explanation of Horizon AI from fragments discovered across the site and LinkedIn, but had to work around a single-page app structure that made content retrieval difficult and incomplete. The final response reveals significant gaps: no transparent pricing was found, competitor positioning is absent, and key product details (supported models, ATP Token specifics) remain unclear or inferred rather than explicitly stated on the site.

  • ›Step [1] (homepage) returned only the HTML skeleton of a Next.js SPA with no content visible in the response; the agent could not extract what Horizon AI does from the raw fetch alone and had to rely on prior knowledge and LinkedIn data.
  • ›Step [4] (about page) similarly returned a bare HTML structure with truncated CSS/JS includes, suggesting the site renders content client-side—making it unusable for agent content extraction without a browser.
  • ›Step [15] (LinkedIn company page) was the only fetch that likely contained substantive text (company description, employee count, mission) and appears to be what actually grounded the agent's final narrative about Horizon AI's identity, team size, and positioning.
  • ›The agent made 7 web searches attempting to find pricing, competitors, and company details; all searches returned link lists with no snippet text that answered the core questions, forcing reliance on prior knowledge (63%) rather than retrieval.
  • ›The site's single-page app design with truncated HTML responses indicates it is fundamentally agent-hostile: no crawlable content, no semantic markup, no fallback for non-browser agents.
  • ›The agent explicitly called out as 'confusing' or 'not found': pricing model (no ranges, requires sales contact), competitor positioning (entirely absent), supported LLM providers (unstated), ATP Token platform mechanics (unclear), and case studies / social proof (no visible examples).
  • ›The agent had to synthesize answers from prior knowledge about Taiwan, KONST partnership mentions, and generic consulting patterns—not from content the site published in a fetchable form.

Want to run your own?

Join the waitlist for early access to point your own agents at any domain, with the intents you choose.

or talk to us about agent readiness →