The agent successfully compiled a comprehensive explanation of Kernel by assembling information from multiple sources, though it encountered significant friction around pricing specifics. Step [3] provided the core value proposition (browser infrastructure for AI agents), steps [5], [7], and [8] contributed competitive positioning and performance claims, but the agent had to work around an opaque pricing page and acknowledge that exact pricing tiers are not clearly published on the site. The site is moderately navigable but requires persistent searching and inference to answer the full brief.

• ›Step [3] (dev.kernel.sh/index.md) was the only content-bearing fetch that clearly articulated the core offering in machine-readable form: 'KERNEL gives web agents cloud-based browser access for fast, scalable automation.' This markdown file is well-structured and directly answers 'what they do' and 'who it's for,' making it the most agent-ready resource on the site.
• ›Steps [5], [7], and [8] (pricing docs, benchmarks, and blog posts) were fetched but returned HTML stubs with truncated payloads that did not render readable content in the trajectory. The agent relied on prior knowledge and web search snippets to fill in the competitive comparison and pricing model details rather than extracting them from the actual page content returned.
• ›The site publishes pricing pages at two URLs (www.kernel.sh/pricing and www.kernel.sh/docs/info/pricing) but neither returned usable content in the agent's fetches. The agent was forced to disclose in its final answer that 'specific tier details and exact pricing are available on their pricing page, though the site doesn't heavily advertise locked-in pricing tiers'—an admission that the pricing information exists but was not successfully retrieved or is poorly presented.
• ›The agent filled significant gaps using prior knowledge (33% of sourced fetches) and web search results (44%), suggesting the site's own pages are either JavaScript-heavy, behind rendering barriers, or lack structured, copy-accessible pricing and feature tables. The markdown index.md was the only directly useful fetch.
• ›The site lacks a dedicated About page (step [9] returned 404), forcing the agent to scatter company context across blog posts and external sources like Y Combinator. This is a navigability friction point for agents seeking foundational identity information.