The agent could not access the provided Vercel development URL (it was login-gated) but successfully pivoted to research Cambio's main public site and third-party sources to answer all four task questions: what they do, who it's for, pricing (not public), and competitive differentiation. The site required external research rather than direct extraction from the development preview.

• ›Steps [1], [4], [5] hit a login wall on the development Vercel URL and returned only HTML boilerplate with no readable content; the agent correctly recognized this was a gated staging deployment and did not waste effort parsing it.
• ›Steps [7] and [9] (web searches) surfaced the main cambio.ai domain and third-party sources (Y Combinator, BusinessWire, Agora Real Estate); step [8] (fetch of cambio.ai) was the only source that delivered usable marketing content, but the response was truncated in the trajectory log.
• ›The agent had to assemble answers from prior knowledge (20%), web search results (20%), and the single successful fetch of the main site (60%); pricing was explicitly noted as 'not publicly disclosed,' confirming the site does not publish it in a machine-readable or easily findable form.
• ›The development URL's git-branch naming convention was correctly identified by the agent as a signal of an internal preview, not a customer-facing resource; this shows good heuristic judgment but also illustrates that the provided URL was not the right target for the task.