The agent successfully assembled a comprehensive explanation of agent-swarm.dev by fetching its markdown documentation directly. The site publishes structured, machine-readable content at /md/* URLs that contained all key information needed to answer what the product does, who it's for, how it's priced, and how it differs from competitors. The agent had to guess the correct URL pattern (/md/index.md, /md/pricing.md, /md/vs/*.md) rather than discovering these through site navigation or links, but once the pattern was identified, retrieval was straightforward and the content was complete.

• ›Step [5] (/md/index.md) returned the core value proposition in machine-readable markdown: lead agent routing work to isolated worker containers with persistent cross-session memory. This single fetch established what the product does.
• ›Step [9] (/md/pricing.md) provided complete pricing tiers (€30–€100/mo cloud, free self-hosted MIT license, 7-day trial) without requiring the agent to parse HTML or reconstruct from fragments—pricing was directly stated in markdown headers.
• ›Steps [10], [13], [14] returned full comparison pages vs. CrewAI, Vercel eve, and Devin, explicitly positioning the product against direct alternatives with clear differentiators (persistent memory, ownership, Docker isolation, recurring work). These were guessed URLs following the /md/vs/*.md pattern, not discovered through links.
• ›The site does NOT surface markdown documentation through HTML navigation or explicit links—the agent had to know or infer the /md/* URL structure. This is a significant UX gap for human visitors but not for agents that can guess standard documentation paths.
• ›The agent cited all five content-bearing steps in its Sources section, confirming it grounded the entire answer in fetched content rather than prior knowledge or inference.
• ›The site's HTML pages ([1], [3], [6]) appear to be Next.js rendered shells without visible text content in the fetch responses—the actual information lives in markdown. This suggests the site is either client-rendered or the markdown is the canonical source and HTML is dynamically generated from it.
• ›The agent correctly identified gaps (review gates undefined, enterprise pricing hidden, detailed vs. pages for 7+ competitors missing) by noting what was NOT in the fetched content, demonstrating honest boundary-setting.