The entity resolution examples are the most concrete thing in here — and the most important. Fraud AI treating the same person as two records isn't a model failure, it's a data foundation failure. The model was doing exactly what it was designed to do. That distinction matters because it means more model investment won't fix it.
The "backlog as risk register" framing is the right posture shift. Every unresolved data quality issue is a live failure point for an autonomous system that will execute on it at machine speed, thousands of times before anyone notices. That's a completely different urgency than a dashboard being slightly wrong.
I write about production AI systems and distributed backends — the infra layer where these data quality decisions compound into retrieval failures and model drift. Worth a subscribe here too.
The entity resolution examples are the most concrete thing in here — and the most important. Fraud AI treating the same person as two records isn't a model failure, it's a data foundation failure. The model was doing exactly what it was designed to do. That distinction matters because it means more model investment won't fix it.
The "backlog as risk register" framing is the right posture shift. Every unresolved data quality issue is a live failure point for an autonomous system that will execute on it at machine speed, thousands of times before anyone notices. That's a completely different urgency than a dashboard being slightly wrong.
I write about production AI systems and distributed backends — the infra layer where these data quality decisions compound into retrieval failures and model drift. Worth a subscribe here too.