Writings
Writings on AI, data, and the organizational realities behind success and failure.
Why Scientific Software Feels Different
Scientific software does more than process data: it produces claims that researchers must decide whether to believe. Trust depends on whether those claims can be reproduced, challenged, and independently evaluated.
The AI System Nobody Is Allowed to Turn Off
AI systems can become difficult to reverse without anyone deciding to make them permanent. As workflows, skills, and procedures reorganize around a system, the fallback process can quietly disappear.
Optimized, Not Improved
AI does not need to confuse an organization about what it wants to improve. It only needs to make the measurable proxy cheap enough to optimize that the proxy becomes the target.
Who Owns the Workflow?
AI can leave accountability with the person making the final decision while moving control upstream into models, thresholds, interfaces, and governance. The question is who actually owns the workflow when those no longer align.
Where the System Ends and Judgment Begins
Clinical AI pushes uncertainty to the edge, where someone has to decide whether the system's answer applies. The question becomes whether the organization has decided who may disagree, what that disagreement means, and what happens afterward.