“A person clicked approve” no longer survives an audit. My work examines what an organization must actually be able to produce before a system earns the authority to act — and the capability required to keep that authority real.
Doctoral research at Virginia Tech is underway. Working papers will be posted here as they develop.
Most AI governance is theater. Effective oversight works like a clearance system — a port authority with zero ambiguity about criteria or authority, not a rubber stamp and not an indefinite hold.
The test: Does a practitioner who gets denied understand exactly what happened, and what to do next?
Read the essay →An effective CoE is a standing enterprise capability, not a department or a project. Its quietest failure mode is the Hollow Bench — automating away the work through which judgment is built.
The test: Are practitioners developing senior judgment, or being hollowed out?
Read the essay →What oversight has to mean when no person can review every action — the agentic transition, and delegating action without losing command of it. From federated defense enclaves to clinical decision support.
What AI procurement and deployment should demand before a system earns authority to act. The same standard lands in bank model risk, FDA-regulated clinical AI, and federal acquisition.
Governance boards as clearance systems, Centers of Excellence as transformation engines, and the build-versus-govern line — the operating-model thinking senior AI roles are hired to supply.
The Hollow Bench and talent debt — building organizations that develop senior judgment instead of automating it away, in healthcare and technology workforces alike.
When does a human’s authority over an AI system stop being real — and when was the last time the organization checked?
My doctoral research examines how organizations maintain the capability to detect degradation in automated systems, contest outputs they can’t fully evaluate, and recover when automation’s normal path fails. The organizing question is when retained human authority becomes ceremonial rather than substantive — and what organizational conditions determine the difference.
The research stays at the unclassified problem-class level: governing agentic AI in high-consequence environments, not any instance of it. Doctoral research at Virginia Tech; working papers will be posted here as they develop.