The Journal
Essays
Long-form writing on AI, philosophy, psychology, and systems thinking.
-
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.
-
Via Negativa in Data Architecture: Remove More, Build Less
Applying via negativa reduced a 34-component data platform to 19, cutting failure points 41% and recovery time from 47 to 12 minutes.
-
When an Agent Lies: AI Hallucination as Ethical Engineering Problem
In a medical information agent, 4.2% of responses contained fabricated information. At 8,000 daily queries, that is 336 potentially harmful outputs per day. Hallucination in high-stakes contexts is an ethical failure.
-
Change Management Is Debugging Your Org OS
Debugging methodology recovered a stalled ERP migration in 11 weeks. Resistance was not irrational. It was accurate feedback about flawed design.
-
The Banality of Algorithmic Harm: Arendt Applied to Engineering
The most dangerous algorithmic harms come not from malicious actors but from competent professionals making reasonable decisions within systems whose cumulative effects no one examines.
-
Designing Systems That Are Auditable by Default
Organizations that build auditability into architecture from day one spend 62% less on compliance remediation. Auditability is an architectural primitive like logging or authentication.
-
Onboarding Reveals Every Knowledge Failure
An onboarding audit found 23 knowledge gaps affecting the entire organization. Fixing them cut time-to-productivity from 47 to 19 days.
-
The Ethics of Data Collection at Scale
Organizations collect 1,400 data points per customer interaction, up from 200 in 2018. The gap between what we can collect and what we should collect is a technical team's responsibility.
-
The ethics of building systems that replace human judgment
-
Privacy-Preserving AI Is a Competitive Advantage
Implementing federated learning and differential privacy cost 18% more but became the selling point in enterprise deals worth $2.1M combined. Privacy is a competitive advantage.