This is exactly how we've designed our PDLSS boundary enforcement at AstraSync. Permissions are set at autonomous and step up and enforceable at runtime. The agent doesn't need to make the decision, the agent simply requests access, the system determines whether that is allowed autonomously, whether a step up approval is required or whether to reject by imposing a hard limit.
This is exactly how we've designed our PDLSS boundary enforcement at AstraSync. Permissions are set at autonomous and step up and enforceable at runtime. The agent doesn't need to make the decision, the agent simply requests access, the system determines whether that is allowed autonomously, whether a step up approval is required or whether to reject by imposing a hard limit.
That's the pattern. The agent requests. The governance layer decides. The boundary decision never touches the reasoning process.