Research notes
Governed agentic AI, in plain language
Short, honest essays on the ideas behind REMORA — why tool-calling agents need an assurance layer, what model confidence is worth in the hardest cases, and how to keep autonomous actions auditable. Grounded in the research, with the caveats kept.
·7 min readGoverned autonomyArchitecture
Why AI Agents Need an Assurance Layer, Not a Better Model
Alignment makes models safer to talk to. It does nothing to decide whether a specific proposed action should execute. That is a different problem — and it needs a different layer.
Read →·6 min readResearch findingCalibration
The Trust Inversion: When Your Most Confident AI Predictions Are the Most Wrong
A confidence score is only useful where it tracks correctness. We found a regime where it does the opposite — and the safe move is to invert the selection rule, not to trust the number.
Read →