AI Glossary

Systems that plan and execute multi-step tasks (often with tools) rather than only generating text responses.

How independently an agent can act. Higher autonomy needs stronger verification, permissions, and rollback paths.

Retrieval-Augmented Generation. The model pulls relevant documents from a source before answering or acting.

When an agent calls functions/APIs like search, databases, or code execution to complete tasks.

Orchestrating triggers, actions, and integrations (often with approvals) to execute repeatable processes.

Measuring quality and reliability using tests, benchmarks, human review, and guardrail metrics like hallucination rate.

When an AI’s output becomes inconsistent over iterative edits, often losing constraints or style requirements.

Interaction to Next Paint. A Core Web Vitals metric that reflects how quickly pages respond to user interactions.