Forgot about that, was mostly thinking about how AI agents with unrestricted permissions would ideally have some external logging and monitoring, so there would be a record of what it touched. A trace has all of the raw information, so some kind of wrapper around that would be useful.