Ramp's public website scored 37/100 on agentability across its homepage, with slightly higher scores on pricing (47/100) and documentation (49/100). This places the site in the lower tier of agent-readiness, indicating that autonomous AI agents will encounter significant friction when attempting to parse content, navigate flows, or execute tasks programmatically.

The audit identified two critical gaps: machine readability scored 0/100, meaning the site lacks structured metadata (JSON-LD schema markup) and has inconsistent heading hierarchies that prevent agents from reliably extracting entities and understanding page structure. Shadow UI avoidance also scored 0/100, suggesting the presence of interface elements or dynamic content that are not properly exposed to programmatic inspection. While defaults (77/100) and chunking (71/100) performed adequately, control mechanisms (22/100), status communication (35/100), and transparency (38/100) all fall short of the baseline needed for reliable agent operation.

These scores reflect a website built primarily for human users, without the structured semantics, predictable state management, or machine-readable audit trails that autonomous agents require to operate safely and effectively.

To improve agentability, Ramp should prioritize the following actionable fixes identified in the audit:

• Implement JSON-LD structured data using schema.org vocabulary to describe primary entities (products, pricing, features) on each page, enabling agents to reliably extract and understand core information.
• Establish a single

  per page and maintain a contiguous heading hierarchy (h1 → h2 → h3) with no skipped levels, allowing agents to parse document structure and navigate content programmatically.

• Wrap all dynamic status updates in ARIA live regions (aria-live, role="status", or role="alert") so agents can monitor state changes and process completion without polling.
• Add labeled pause and cancel controls for long-running operations (imports, bulk actions) with accessible names like "Cancel import" rather than unlabeled icons, giving agents explicit termination points.
• Provide undo mechanisms for destructive or irreversible actions, and allow agents to override or edit completed results before they take effect.
• Expose a machine-readable activity log or audit trail (visible UI feed backed by JSON event data) that records system actions and includes a "why this result" affordance agents can query to understand decisions.