Essays

AI Ethics in Content Moderation: The Impossible Standard

AI content moderation achieves 92-96% accuracy for clear violations but drops to 54-68% for content requiring cultural context or nuanced judgment. The gap defines an impossible standard.

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.

The Productivity Placebo: METR’s AI Coding Study

METR's randomized trial found AI tools made experienced developers 19% slower while they believed they were 24% faster. The perception gap demands investigation.

The Art of the Escalation: When and How to Ask for Help

Delayed escalations cost an average of 4.2 additional engineering days per incident. The median delay between knowing help was needed and asking for it was 2.3 days.

The paradox of AI-assisted creativity—can tools that compress knowledge also expand imagination?

Integration Architecture Is Where Good Systems Go to Die

Integration points accounted for 58% of production incidents despite representing less than 15% of total codebase. Integration is where architectural quality is tested most severely.

The difference between systems and processes (and why it matters for your work)

AI Ethics in the Supply Chain: Training Data Provenance Problem

Tracing training data lineage for 3 models revealed none could document full provenance. One model included 12 sources with no consent chain. AI has a data supply chain problem.

Jira Has a Philosophy Problem, Not a UX Problem

Jira attempts to be both a coordination tool and a surveillance instrument. Separating these functions reduced cycle time by 28%.

Estimation Is Epistemology: Confidence Intervals

Calibrated teams delivered within 15% of estimates 74% of the time versus 31% for gut-feel. The difference was self-knowledge, not technical skill.