Watch how an agent would navigate your site.

Pick a domain and a task. We simulate the journey an AI agent (Claude, ChatGPT, Perplexity, Operator) would take: fetch, parse, look for relevant links, hit blockers, extract content, and either complete the task or get stuck.

simulated walk Steps are derived from the latest public scan’s data (robots.txt rules, JS-required signal, cookie walls, sitemap, JSON-LD presence), not a live LLM run.

1. A numbered step-by-step journey: fetch, parse, look for affordances, navigate, extract, outcome.
2. Failure points highlighted: cookie walls, login walls, JS-required content, performance timeouts, sparse text-to-DOM ratio.
3. A short list of suggested fixes, each linked to a copy-pastable remediation on /score/<domain>/fix/<check>.

v1 uses static analysis over the latest scan’s data. It does not execute the four predefined tasks live. There are no real Anthropic or Browserbase calls, no clicks, no form submissions. The walk is a plausible reconstruction of what an agent would face based on what AgentSpeed already measured. Live LLM-driven runs ship with the monitoring tier.