Glossary
AI agent glossary
The words that come up while evaluating agent platforms, defined in one sentence each and then explained. Written for someone making a decision rather than someone writing a paper — so each entry leads with what the term means and follows with why it matters to you.
The basics
What the category is, and the unit inside it.
Memory and learning
What an agent keeps, and what actually changes over time.
Agent memory
What an agent retains between sessions, stored outside the model rather than in its weights.
Agent skill
A written procedure followed the same way each time, captured once rather than re-explained.
RAG
Searching your documents for relevant passages and giving them to the model before it answers.
Working together
What happens once there is more than one agent.
Control and trust
The vocabulary of keeping an agent inside bounds you chose.
How to use this
If you are evaluating platforms, the four terms that most often separate a serious product from a demo are guardrails, human in the loop, autonomy levels and agent memory. A vendor who is vague on all four has built a chatbot with ambition.
For the longer versions, the pillars: AI agents, multi-agent platforms, self-improving agents and AI for business.
See the terms in something running
Most of these are easier to understand in a live deployment than on a definitions page.