The Journal
Essays
Long-form writing on AI, philosophy, psychology, and systems thinking.
-
Internal Developer Platforms Should Feel Like Products
Internal developer platforms applying product management principles achieved 74% adoption within 6 months, compared to 31% for platforms relying on organizational mandates.
-
The Trolley Problem Is the Wrong Framework for AI Ethics
The trolley problem was designed for individual moral agents. AI systems are sociotechnical institutions. Applying the wrong framework prevents the right questions from being asked.
-
Your Data Catalog Is Lying to You
An audit of 3 enterprise data catalogs found that 38% of table descriptions were inaccurate and 22% of documented columns no longer existed. Catalogs create false confidence.
-
AI Ethics Guidelines Are Architecture Requirements
Treating AI ethics guidelines as architecture requirements reduced post-deployment ethical incidents by 67% across 4 production systems. Ethics constraints force better engineering discipline.
-
Observability for Ethical Systems: Monitoring Beyond Uptime
Adding fairness metrics to a recommendation system serving 1.2 million users revealed 3 systematic biases that traditional uptime and latency monitoring missed over 14 months.
-
The Vibe Coding Trap: Engineering Fundamentals With AI
Vibe-coded projects had 4.1x more production defects than AI-augmented engineering projects. AI tools make engineering judgment the primary bottleneck.
-
The Architecture of Trust: Designing Systems People Can Rely On
Systems rated highly trustworthy share 4 architectural properties: predictable failure behavior, transparent state reporting, bounded response times, and honest error messages. All require deliberate design.
-
Onboarding as Knowledge Architecture
Redesigning onboarding as a knowledge architecture problem reduced new hire time-to-productivity from 14 weeks to 8 weeks and 6-month attrition from 22% to 9%.
-
On Trusting Systems You Cannot Fully Inspect
Deploying opaque AI systems creates tension between utility and epistemic responsibility. Current interpretability explains roughly 30-40% of model behavior.
-
Prompt Patterns as Architectural Contracts
Treating prompt templates as versioned architectural contracts reduced production incidents by 72%. Prompts deserve the same engineering rigor as API specifications.