The agent pieced together a functional understanding of Synth AI's core purpose (automated training for AI coding agents) and positioning, but could not access the pricing page or complete the full feature picture due to client-side rendering. It successfully used the Y Combinator launch page [15] and docs SDK overview [12] as primary sources, supplemented by web search fragments, but the main website remained largely opaque.

• ›Step [15] (Y Combinator launch page) was the single most content-bearing source—it contained the core value proposition, the SWE-Bench benchmark improvement claim (33% relative improvement), and product positioning. This was cited in the final response's sources.
• ›Step [12] (docs.usesynth.ai/sdk/overview) loaded with Astro HTML but the agent could not extract readable content from it due to truncation; the agent inferred SDK integration patterns from search fragments and prior knowledge rather than from what step [12] actually returned.
• ›Steps [1], [3], [4] (the main website and /pricing, /docs routes) returned Next.js client-side rendering boilerplate with no readable HTML content—the agent could not extract any information from them and had to work around them.
• ›The pricing page [3] was unreachable in practice (returns client-side scaffold only), and the agent correctly flagged this as a critical gap. No search results surfaced accessible pricing information.
• ›The agent relied heavily on web search [9, 11, 13–21] to disambiguate between Synth AI and similarly-named competitors (Synthesia, Synthflow, etc.), but search results were noisy and did not directly answer pricing or detailed feature questions.
• ›The site does not publish machine-readable positioning, pricing, or feature data; the agent had to assemble a narrative from a single external source (YC launch) and infer the rest from SDK naming conventions and context clues.