A/B test

An A/B test is an experiment that compares two versions of a system, such as two prompts or models, by exposing different groups of users to each and measuring which performs better on chosen metrics.

A/B test sits in the Evaluation & safety part of the Agentik {OS} glossary, which defines the words used to build and run AI agent systems.

Also called A/B testing, Split testing.