How agent-ready is Salesforce?

Salesforce's agentability audit returned an overall score of 27/100 on the homepage, 20/100 on pricing, and 38/100 on documentation. These scores indicate significant barriers for autonomous AI agents attempting to parse, navigate, and operate the platform's public web presence. Agent-readiness remains limited across most measured dimensions.

The platform performed moderately on chunking (65/100) and machine readability (50/100), suggesting content is somewhat structured for programmatic consumption. However, Salesforce scored zero on three critical principles: control, status, and transparency. This means agents have no visibility into system actions, cannot reliably manage long-running operations, and lack mechanisms to verify or reverse destructive changes. Shadow UI avoidance (30/100) and clean handoffs (23/100) also scored below functional thresholds.

These gaps primarily affect agents' ability to safely execute workflows, understand system state, and recover from errors—core requirements for autonomous operation at scale.

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

• Expose an activity or audit log that records system actions in machine-readable form, such as a visible activity feed paired with a JSON event log, to enable transparency into what the system has done.
• Add JSON-LD structured data using schema.org vocabulary to describe primary entities on each page, improving machine readability for agents parsing content.
• Gate destructive actions behind explicit, distinctly-labelled confirmation steps rather than same-styled buttons, and support undo or override capabilities on completed actions to give agents safe control over workflows.
• Provide labelled pause, cancel, or stop controls for long-running operations with accessible names like 'Cancel import' instead of unlabelled icons, so agents can manage processes in flight.
• Surface confidence or certainty metadata on key outputs—such as a confidence field in API responses or 'verified vs best-guess' labels in the UI—to help agents assess result reliability.
• Attach source references or input citations to generated outputs so agents can trace the provenance of data and verify accuracy.