Agent Identities

Agent Identities examines how artificial intelligence agents acquire, constrain, and exercise authority within enterprise systems, comparing autonomous machine identities governed by service principals against delegated per-user On-Behalf-Of (OBO) token exchanges. Rooted in production implementations across Salesforce, Microsoft Entra ID, Jira, and Anthropic's Model Context Protocol (MCP), the book demonstrates why identity architecture dictates enterprise security boundaries. By dissecting two real-world systems—the autonomous ticket summarizer (sfsummary) and the interactive multi-client MCP gateway (mcp-client)—alongside deep dives into RFC 6749, RFC 7523, RFC 8693, and RFC 9728, this architectural investigation establishes that an AI assistant in an enterprise environment must never possess more power than the authenticated human invoking it.