Definition
What it means
Data annotation turns raw or generated material into examples that an AI system can learn from or be measured against. Depending on the task, annotators may classify content, write demonstrations, compare responses, apply a rubric, correct an output, or review a sequence of actions.
Why it matters
Role in the system
Annotation makes human and domain expertise explicit enough to guide training and evaluation, while annotation quality directly affects the reliability of the resulting signal.
Example
In practice
A software engineer reviews an agent's terminal session, marks whether each action is valid, and records a corrected solution for supervised fine-tuning.