Robert R. Dogonowski
AI governance · verification · institutional economics
Independent research on how organisations — and the individuals inside them — can trust AI outputs in consequential work. Working across the economics of verification, the limits of agency in artificial agents, and governance architectures for high-liability AI deployment.
Based in Copenhagen. The work spans organisational economics, contractual inference, and the behavioural infrastructure of solo AI use. Selected output appears below; pre-prints and working papers are available on SSRN.
Research areas
The economics of AI verification
A microeconomic treatment of why solo and institutional AI users systematically over- or under-verify outputs, and what the optimal allocation of verification effort looks like under explicit liability decomposition. Published as The Solo AI Verification Operating System with the Cost Floor Theorem as its analytical core.
Agency without internal discipline
Classical principal-agent theory presumes internalised consequences — reputation, career, intertemporal utility — that AI agents lack. The research programme asks what governance structures replace internal discipline when delegating cognitive work to artificial agents, and how organisations should adapt classical agency contracts accordingly.
Contractual inference
A framework for institutional governance of AI reasoning. Separates intent, execution, and verification across a Governor-Generator architecture in which contractual rules rather than capability improvements bound the system's behaviour. Addresses the deployment problem in regulated industries where AI capability outpaces verification cost.
The Steering Wheel Prompt Protocol
A process-level protocol for AI reliability under uncertainty. Model-agnostic and tool-agnostic. Defines the failure modes that cause unreliable outputs even when underlying models are capable, and specifies a deployment pattern that treats reliability as a process problem rather than a capability problem.
The competence trap
An observation that verification cost can increase counterintuitively with AI capability improvements. As models become more capable, residual errors become subtler and harder to detect, raising the effective cost of trustworthy adoption in high-stakes contexts. The mechanism creates structural barriers to AI deployment in regulated industries.