Latest data news

Data platforms

Google builds synthetic test estates for data agents

Google says its internal simulator lets teams test enterprise data agents against realistic synthetic company data without exposing customer information.

What happened

The Google Cloud Blog describes a simulator that generates linked databases, documents and chat histories from a short description of organisational patterns. A frozen blueprint and random seed produce repeatable data and verified answers for agent tests.

Why it matters

The approach is intended to help engineering teams test agents without access to sensitive production data or hand-labelled benchmarks. It also introduces realistic conflicts and errors that clean test databases may miss.

What to watch

Teams could check whether agent rankings and accuracy changes in synthetic tests match results on human-authored benchmarks; Google says broader paired-estate studies are under way.