Salesforce technical architect Scott Ratliff published a guide on June 22, 2026, detailing how to implement Salesforce Headless 360 to connect external AI agents to core customer data securely. The scenario centers on an AI-native startup requiring real-time investment recommendations while maintaining strict governance standards. Architects are encouraged to follow a structured three-step methodology—What, How, and Why—to move from business requirements to documented architectural decisions.

The "What" phase identifies five High-Level Requirements (HLRs): ingesting market signals, retrieving client mandates from Salesforce, evaluating thesis compliance, surfacing recommendations to advisors, and escalating issues to human agents. For the "How" phase, the design avoids using an Agentforce agent for data retrieval, instead opting for a Model Context Protocol (MCP) server. To ensure security, the architecture employs named queries to restrict agent access to specific fields and objects, rather than granting broad read permissions. User context is utilized for MCP server connections to enforce client-advisor data boundaries.

The "Why" phase emphasizes the use of Architectural Decision Records (ADRs) to link design choices with underlying assumptions. According to the guide, traditional APIs act like basic phonebooks, whereas MCP servers function as semantic toolkits that provide agents with specific schemas and instructions. Salesforce Headless 360 is recommended specifically for organizations that prefer to keep proprietary AI in established interfaces rather than moving operations into the Salesforce platform. Ultimately, the system design prioritizes granular security by employing custom MCP servers with invocable actions to maintain the principle of least privilege during compliance escalations.

Salesforce technical architect Scott Ratliff published a guide on June 22, 2026, detailing how to implement Salesforce Headless 360 to connect external AI agents to core customer data securely. The scenario centers on an AI-native startup requiring real-time investment recommendations while maintaining strict governance standards. Architects are encouraged to follow a structured three-step methodology—What, How, and Why—to move from business requirements to documented architectural decisions.

The "What" phase identifies five High-Level Requirements (HLRs): ingesting market signals, retrieving client mandates from Salesforce, evaluating thesis compliance, surfacing recommendations to advisors, and escalating issues to human agents. For the "How" phase, the design avoids using an Agentforce agent for data retrieval, instead opting for a Model Context Protocol (MCP) server. To ensure security, the architecture employs named queries to restrict agent access to specific fields and objects, rather than granting broad read permissions. User context is utilized for MCP server connections to enforce client-advisor data boundaries.

The "Why" phase emphasizes the use of Architectural Decision Records (ADRs) to link design choices with underlying assumptions. According to the guide, traditional APIs act like basic phonebooks, whereas MCP servers function as semantic toolkits that provide agents with specific schemas and instructions. Salesforce Headless 360 is recommended specifically for organizations that prefer to keep proprietary AI in established interfaces rather than moving operations into the Salesforce platform. Ultimately, the system design prioritizes granular security by employing custom MCP servers with invocable actions to maintain the principle of least privilege during compliance escalations.