How we think
Long-form, practical, opinionated. No newsletter pop-ups. RSS if you want it.
Most recent first.
Trust is architecture, not a feature
Why trustworthy AI products need product architecture, not just better AI copy, explainability labels, or governance theatre.
Approval fatigue: why human-in-the-loop fails at scale
Human-in-the-loop AI fails when approval is overused, under-contextualized, and poorly placed. Here is how to design oversight that survives real workloads.
How to audit an AI product in 10 days: the five lenses
A five-lens AI product audit method for finding trust, oversight, and recoverability problems in AI features before they become adoption problems.
Recoverable AI: designing systems that fail safely
Recoverable AI is a product architecture approach for designing AI systems that can be corrected, reversed, and trusted even when they fail.
The agent has a plan. Can the user see it?
Planning visibility is a core UX requirement for AI agents. Users need to see, edit, and approve agent plans before consequential work happens.
Shadow AI is a product-design failure
Shadow AI is not only a governance problem. It is evidence that official AI workflows are too slow, unclear, locked down, or badly fitted to real work.
The AI action ledger: making autonomous work attributable
AI action ledgers make autonomous work inspectable and attributable by recording plans, actions, approvals, sources, and recovery paths.
AI governance belongs in the workflow, not the PDF
AI governance becomes real when it is designed into workflows: permissions, approval gates, previews, logs, escalation paths, and recovery mechanisms.