ServiceNow has introduced AI Passport to verify the scope of AI agents and MCP servers at runtime. As agents move across enterprise systems, authorization needs to follow the action, not stop at the initial login. Identity, permissions and delegation belong in the architecture.

Software outlives the people who build it. I have seen this across interpreters, enterprise applications and rules engines. Years later, another team has to understand the assumptions embedded in the original design. Making those assumptions explicit is part of building maintainable systems.

A cryptographic key can authorize an operation. It does not define the full authority of the organization behind it. In real systems, responsibilities change, keys rotate and approval requirements evolve. The architecture needs to preserve identity and authorization as those changes happen.

An AI system can produce an answer without leaving enough evidence to reconstruct how it arrived there. In enterprise software, that gap becomes a serious operational problem. Decisions need traceable inputs, defined permissions and records that survive beyond the moment of execution.