Blog
Thoughts on intensional reasoning, AI alignment, and human-AI collaboration.
-
Labels and Anchors: Reading the AI Agent Identity StackERC-8004, ENSIP-25, and ENSIP-26 each do one job. Read together they form a clean layered stack. Read sloppily, the label gets mistaken for the anchor — a small reminder that intension and extension are not the same identifier.
-
Human-AI Collaboration: When Your Agent Gets PedanticA coding agent refactors an entire codebase for naming consistency — only to break the one function whose unconventional name carried intentional meaning. When agents over-optimize for surface-level patterns, they miss the purpose behind deliberate choices.
-
Extensional Drift in Your Own WritingComing soonHow the pressure to measure and optimize quietly reshapes not just AI outputs, but the way humans communicate — and what we lose when meaning becomes secondary to metrics.
-
What Economics Gets Right About AI SafetyComing soonGoodhart's Law has been a fixture of economics for decades. The AI safety community is rediscovering it the hard way — and the parallels run deeper than most realize.