AI-agent readiness is a task outcome.

AI-agent website readiness is the ability of a declared agent profile to complete a representative customer goal on a specific website and environment. The result must be verified independently from the agent's own final message.

The scope matters. A public research task, an authenticated account change and a sandbox checkout have different permissions, data and stop gates. A website can be ready for one task and fragile for another, so a single universal pass label hides useful information.

Aiscovery separates readiness into three surfaces so teams can locate the actual failure without treating every technology signal as a product outcome.

A useful scenario starts with a persona, goal, prerequisites and controlled test data. It declares allowed actions, the irreversible action where production must stop and a deterministic assertion that proves the final state.

For an e-commerce journey, the goal may be to find a compatible product, confirm delivery terms and reach cart review without placing an order. For SaaS, it may be to find a sourced answer and reach the first safe signup step. The contract stays stable while agents and models change.

A success message from an agent is not enough. Verify the final URL, visible state, network response or sandbox record independently. Preserve screenshots, step trace, accessibility snapshots and relevant console or network evidence around the point of failure.

Track completion time, retries, steps, interventions and cost beside the outcome. Measure Cumulative Layout Shift and target movement around important actions because an agent can identify the correct control and still actuate the wrong location after late content moves the page.

A website defect, agent-specific limitation, policy block and test-infrastructure failure require different owners and remedies. Re-run the same versioned task with a deterministic browser baseline and multiple agent profiles before assigning the cause.

If one agent fails while the baseline and other agents succeed, preserve the exact decision, tool call or policy boundary that diverged. If every executor fails at the same rendered state, the evidence points back to the website or shared environment.

Semantic HTML, accessibility, Schema.org, llms.txt, OpenAPI, MCP and WebMCP can improve particular surfaces. Their mere presence does not prove that a customer task works, and the absence of an emerging interface should not automatically fail a site with a reliable accessible path.

Chrome's Agentic Browsing category currently presents deterministic checks as actionable signals rather than a definitive zero-to-one-hundred ranking. Use the same discipline in a readiness audit: record lifecycle and conformance separately, then connect each finding to an observed task impact.