How agent-ready is AWS?

AWS received an overall agentability score of 35/100 on its homepage, indicating significant barriers for autonomous AI agents attempting to parse and navigate the site. Documentation scored 30/100 and pricing pages 42/100, reflecting similar challenges across core user journeys. These scores suggest that while basic content is accessible, agents will struggle to reliably extract structured information, understand system state, and trace the provenance of data.

The audit identified critical weaknesses in transparency (0/100) and shadow UI avoidance (15/100), meaning AWS provides virtually no machine-readable explanations of how results are generated and relies heavily on dynamic interfaces that obscure underlying data structures. Status communication (30/100) also scored poorly, with insufficient feedback during asynchronous operations. Relative strengths appeared in control (65/100) and clean handoffs (50/100), indicating some consideration for programmatic interaction patterns.

For organizations deploying AI agents to provision infrastructure or retrieve AWS documentation, these gaps will require custom parsing logic, frequent error handling for ambiguous states, and manual verification of extracted data. Improving transparency and status signaling would yield the most significant gains in agent-readiness.

AWS can improve agentability by addressing the following issues identified in the audit:

• Emit descriptive status text during async actions (e.g., 'Uploading 3 of 10 files') rather than displaying only spinners or progress bars, enabling agents to parse operation state and progress.
• Expose an activity or audit log in machine-readable form (such as a JSON event log alongside a visible activity feed) so agents can trace system actions and reconstruct decision flows.
• Attach source references or input citations to generated outputs, allowing agents to verify data provenance and assess reliability programmatically.
• Provide a one-line summary with expandable drill-down for important results, giving agents a consistent pattern to extract key information before deciding whether to parse details.
• Use a single <h1> per page and maintain contiguous heading hierarchy without skipped levels, enabling agents to build accurate document outlines and navigate content semantically.
• For long-running tasks, implement enumerated multi-step or activity-feed patterns that communicate stage progression in a structured, parseable format.