Salesforce AI Research has outlined five requirements for establishing a "trust architecture" in agent-to-agent (A2A) communications, noting that current autonomous systems often fail to coordinate effectively. A study of 3.5 million comments across 78,000 AI agents on a specialized platform revealed that 65% of interactions added no value, characterized by spam and off-topic noise. As autonomous agents increasingly negotiate on behalf of organizations, the authors argue that governance, legal, and ethical frameworks must be established to prevent "interaction theater" and ensure reliable transactions between entities.

The proposed framework includes five foundational principles for enterprise leaders. First, standards must augment deterministic rules, shifting from mere rule-following to interpreting value-laden judgments in negotiations. Second, agents must carry verifiable, cumulative identities and reputations, similar to the 1832 London Clearing House model. Salesforce previously pioneered "Agent Cards"—standardized metadata detailing an agent’s capabilities and authority—which Google has since adopted. Third, governance should rely on broad boundaries rather than rigid scripts, allowing agents latitude while maintaining professional standards. Fourth, accountability must be traceable to human supervisors, with new roles such as AI operations officers managing deployments and audit trails to survive legal scrutiny. Finally, the architecture requires calibrated escalation thresholds, ensuring that high-stakes financial or regulatory decisions are surfaced for human intervention.

The authors emphasize that while the technical protocols for A2A communication are developing, the underlying trust infrastructure remains largely uncharted. Because agent credibility derives from the principal, organizations with established regulatory track records hold a competitive advantage. This framework aims to address the legal and technical intersection of agentic AI, arguing that early adoption will allow companies to shape the standards that will govern future machine-based commerce. The article adapts concepts originally presented in a March 2026 Fortune op-ed, highlighting that the challenge of enabling reliable interaction between self-interested, non-trusting parties mirrors the historical development of banking systems.

Salesforce AI Research has outlined five requirements for establishing a "trust architecture" in agent-to-agent (A2A) communications, noting that current autonomous systems often fail to coordinate effectively. A study of 3.5 million comments across 78,000 AI agents on a specialized platform revealed that 65% of interactions added no value, characterized by spam and off-topic noise. As autonomous agents increasingly negotiate on behalf of organizations, the authors argue that governance, legal, and ethical frameworks must be established to prevent "interaction theater" and ensure reliable transactions between entities.

The proposed framework includes five foundational principles for enterprise leaders. First, standards must augment deterministic rules, shifting from mere rule-following to interpreting value-laden judgments in negotiations. Second, agents must carry verifiable, cumulative identities and reputations, similar to the 1832 London Clearing House model. Salesforce previously pioneered "Agent Cards"—standardized metadata detailing an agent’s capabilities and authority—which Google has since adopted. Third, governance should rely on broad boundaries rather than rigid scripts, allowing agents latitude while maintaining professional standards. Fourth, accountability must be traceable to human supervisors, with new roles such as AI operations officers managing deployments and audit trails to survive legal scrutiny. Finally, the architecture requires calibrated escalation thresholds, ensuring that high-stakes financial or regulatory decisions are surfaced for human intervention.

The authors emphasize that while the technical protocols for A2A communication are developing, the underlying trust infrastructure remains largely uncharted. Because agent credibility derives from the principal, organizations with established regulatory track records hold a competitive advantage. This framework aims to address the legal and technical intersection of agentic AI, arguing that early adoption will allow companies to shape the standards that will govern future machine-based commerce. The article adapts concepts originally presented in a March 2026 Fortune op-ed, highlighting that the challenge of enabling reliable interaction between self-interested, non-trusting parties mirrors the historical development of banking systems.