The Journal
Essays
Long-form writing on AI, philosophy, psychology, and systems thinking.
-
Bias Detection Tools Are Only as Good as Your Data Model
Bias detection tools identified only 34% of bias issues I encountered in production. The other 66% originated in data modeling decisions upstream of any tool's reach.
-
Autonomous Agents Need Epistemology, Not Parameters
The bottleneck in autonomous agents is epistemic competence, not capability. Implementing uncertainty quantification reduced confidently wrong actions by 78%.
-
The Data Warehouse Is Not Dead; Your Expectations Were Wrong
After migrating to a lakehouse and back, a company reduced query costs by 52% and improved report reliability from 87% to 99.1%. The warehouse is not dead. It was misunderstood.
-
AI in Research Has a Reproducibility Problem Ethics Frameworks Ignore
In 47 AI-assisted research papers, 62% treated model outputs as findings and 72% lacked reproducibility documentation. AI amplifies the replication crisis when ethics frameworks ignore methodology.
-
Vanity Metrics and the Theater of Data-Driven Decision Making
Only 4 of 15 self-identified data-driven organizations demonstrated a decision changed by data last quarter. The rest perform data-drivenness without practicing it.
-
Designing for Composability: Building Systems From Interchangeable Parts
Systems designed for composability reduced feature development time by 41% and integration effort for new channels by 67% across 9 organizations adopting composable architecture principles.
-
The Ground Truth Problem: When Your Labels Are Wrong
A review of labeling accuracy in 5 production ML datasets found error rates from 4% to 17%. When labels are wrong, models faithfully reproduce human mistakes at scale.
-
Data Observability Needs Product Thinking
Standard observability caught 78% of structural failures but only 11% of semantic failures. The undetected semantic failures caused 19x more business impact.
-
Open Source AI Ethics: Who Governs Models Without Owners
Open-weight models downloaded over 500 million times have no single entity controlling deployment. When harm occurs, the question of responsibility has no clear answer in current frameworks.
-
Feedback Cultures Are Built, Not Declared
78% of engineers reported withholding critical feedback at organizations that declared feedback culture but lacked structural mechanisms. Feedback requires infrastructure.