Insights.
Research notes on software, artificial intelligence, interaction and the craft of building original products.
Design AI Interfaces Around Uncertainty
A practical four-band method separates evidence quality from interface permission, so AI uncertainty leads to proportionate human-directed action.
Build an AI Creative-Direction Regression Set
Test more than technical validity: build a compact evidence set that shows whether an AI workflow still serves its intended creative direction.
How to Audit AI Without Logging Sensitive Data
A practical field-by-field method for preserving useful AI operational evidence without turning observability into a hidden store of user content.
Archive
How to Design Human Review Controls for AI
A practical control ladder helps teams give people the right way to inspect, reject or replace an AI-assisted result.
Data Minimization for Useful AI Features
A practical checklist for giving an AI feature the context it needs without exposing every available record or collecting unnecessary personal information.
Least-Privilege Permissions for AI Tools
A practical framework for giving a tool-using AI capability only the information and actions required for its intended product role.
AI in financial software: what should remain deterministic?
A practical Viotus framework for separating financial data, deterministic engines, AI assistance and the decisions that remain human.