OMB put AI governance on the calendar. The data definition still needs a chair.
A board can convene the right people. It still has to make one of them responsible.
Every entry, filed by the failure mode it exposes. Nothing here is deleted — only filed.
A board can convene the right people. It still has to make one of them responsible.
A management system can preserve discipline. It cannot supply the missing vocabulary.
Common names are valuable. They are not the same as common accountability.
Gartner brings back its MDM map. The single view it charts has been under construction the whole time.
Salesforce closed its purchase of Informatica. The master-data vendor is now master data.
A right to access connected-product data arrives before anyone agrees what the fields mean.
A privacy-enhancing technology whose privacy is, by default, off.
A compatibility check can block a breaking change. It can't name who's allowed to make one.
Debezium 3.0 ships a connector that refuses to guess. The org still guesses.
Consumption pricing was a feature until the invoice became the strategy.
The keep-everything reflex met a hard-drive shortage, and the reality tax came due.
The EU AI Act's data-governance clause assumes lineage, provenance, and bias records most teams were never resourced to keep.
Europe's first big AI-training fine got voided. Not for being wrong — for being orphaned.
A real standard for data contracts now exists. The argument it was supposed to settle has simply moved up a layer.
The catalog logs in. The steward logs the catalog. Nobody logs the decision the purchase was meant to replace.
OpenAI's December outage was caused by the service meant to watch for outages.
An industry consortium standardized the file format for metric definitions. The meeting where Finance and Sales disagree is still on your calendar.
The Generative AI Profile treats provenance as a control — but admits most builders cannot say what they trained on.
dbt moved data contracts into the build step. It also lets you declare keys the warehouse never actually checks.
dbt open-sourced MetricFlow so agents query one definition. Whichever definition wins by default is now the one the robots inherit.
Anomaly detection now defines 'good' for you. It defines it as 'whatever usually happens.'
Airflow and dbt will draw your pipeline for nothing. The arrow still dies one hop short of the meeting where the number gets used.
The FTC's fix for an ed-tech breach wasn't more storage. It was a retention schedule someone has to own.
An engineering convenience became a control point. A control point needs a controller.
The obligation assumes an inventory the organization skipped. The inventory is the project.
The catalog filled with products. The org chart did not move an inch.
A define-once layer is an admission that the term was never defined.
The catalog scanned every table. It has not been opened since the rollout.
You can monitor a metric to the second and still not know what it counts.
The models needed one definition of "customer." That was always the assignment.
The retriever inherits every undefined term in the corpus. The model just reads it aloud.
A graph that stops at the warehouse door explains everything except where the number came from.
Minimization is a sentence about purpose. Most firms never finished the sentence.
The role turns over in months. The undefined terms have been here for years.
Storage is interoperable now. The decisions about it are still proprietary.