The agent successfully assembled a comprehensive explanation of Zed by combining prior knowledge with web search results, since the zed.dev website itself was too heavily client-side rendered to extract content directly. The final answer covered what Zed does, its target audience, pricing structure, and differentiation from competitors—all accurate and detailed—but relied on external sources and educated guessing rather than on-site documentation being machine-readable or well-structured.

• ›Step [1] (homepage) and [3] (pricing page) returned only HTML scaffolding with no extractable text—both were client-side rendered Next.js apps that provided no server-rendered content to the agent. The agent was unable to read the site natively and had to pivot to search.
• ›Steps [6–9] (web searches) were the true content sources: search result snippets and linked third-party review sites (CostBench, CompEdge, Tech Insider, DevToolReviews) provided the performance metrics, pricing tiers, feature comparisons, and positioning that filled the final answer. The agent never successfully fetched and parsed zed.dev's own pricing or comparison pages.
• ›The agent accurately cited zed.dev URLs in its Sources section (pricing page, AI page, compare/vscode page, docs) but these were not the actual sources of the information—they were reconstructed from prior knowledge and search result links. The agent cited pages it could not read, suggesting it inferred their existence and relevance rather than extracting from them.
• ›Critical gaps acknowledged by the agent: extension ecosystem availability, AI model pricing opacity, enterprise feature specifics, free tier edge cases, and platform support details—these are all items that should be documented on-site but either weren't fetched or weren't discoverable because the site is not agent-friendly.
• ›The site's heavy reliance on client-side rendering (Next.js) made it invisible to the agent's direct fetch strategy. This is a significant friction point for AI agents: real content (pricing, comparisons, feature lists) was locked behind JavaScript rather than available as static HTML or structured data.