The agent successfully synthesized a comprehensive explanation of Axiom.co by working around a JavaScript-heavy website that returned unreadable HTML. It relied on web search results and documentation links (steps 7, 9, 10, 13) plus the fetched docs introduction (step 8) to piece together what Axiom does, who it's for, pricing, and competitive positioning. The site itself was poorly navigable for automated agents due to client-side rendering, forcing the agent to escape to search and third-party sources for the core answer.

• ›Steps [1], [3], [4], [5], [11], [12] all returned identical truncated HTML boilerplate with no readable content—the site's React rendering made direct page fetches useless. The agent had to abandon these and pivot to search.
• ›Steps [7], [9], [10], [13] (web searches) and step [8] (docs introduction fetch) were the only content-bearing sources. Search result snippets and the docs page provided enough structured text to build the answer. The agent explicitly cited 8 URLs in its Sources section, 5 of which came from search discovery or docs—not from crawling the main site.
• ›The agent noted multiple critical blockers: (1) main website is JS-rendered and opaque to crawlers, (2) exact pricing details ('cost per GB at different volume levels') are not published in machine-readable form, (3) pricing page returned unreadable HTML despite being fetched twice, (4) feature/product pages were similarly inaccessible. The agent had to infer pricing from third-party aggregators (PricingSaaS) rather than Axiom's own pricing page.
• ›The agent correctly identified that Axiom.co owns a 'schema-on-read' architecture, APL query language, and 25-50x compression—but these specifics came from search snippets and docs, not from the website's own public pages. The site does not expose its key differentiators in a machine-readable layer (no structured data, no plain-text docs on main pages).
• ›The agent's final answer was well-formed and comprehensive, but required 43% of fetches to come from 'previous_artifact' (likely cached docs or search results), 29% from web search, and only 29% from prior knowledge—indicating heavy reliance on external data bridges rather than the site's own navigation.