The agent successfully gathered a comprehensive explanation of Jo by working around a poorly navigable website. The main site lacked a pricing page and detailed service descriptions, forcing the agent to rely on the FAQ and especially the /llms.txt file—a machine-readable summary explicitly written for LLMs—plus web search results. The site's agent-readiness was ultimately good due to the /llms.txt artifact, but discoverability of pricing and core positioning was weak on the marketing pages themselves.

• ›Steps [6] and [7] were the only content-bearing sources that answered the task substantively. Step [6] (FAQ) was found through web search and contained structured Q&A; step [7] (/llms.txt) was discovered via a comment in step [6] and provided the canonical structured definition of Jo's architecture, capabilities, and positioning.
• ›The site actively failed to surface pricing and company info through normal navigation: /pricing returned 404, /about returned 404, and the homepage (step [1]) did not contain pricing or detailed capability descriptions in its HTML. The agent had to detour through web search (steps [4]–[5]) to find third-party sources and the FAQ URL.
• ›The /llms.txt file was explicitly designed as 'the canonical, structured source' for LLM understanding, indicating Jo anticipated AI agents would need to fetch the site and provided a machine-readable escape hatch. This is high agent-readiness, but it also signals the marketing site itself was not structured for agent consumption—the company felt obliged to create a separate document to compensate.
• ›The agent had to assemble a complete answer from fragments: pricing from web search results (not the site), capabilities from /llms.txt, use cases and differentiation from third-party profiles and the /llms.txt file, and limitations (Mac hardware requirements, iOS timeline) from scattered sources rather than a unified product page.