The basics
- AI agent. Software that completes a task by taking several steps, such as reading a message, checking a calendar and booking a slot.
- Chatbot. Software that answers questions in a conversation. See agent vs chatbot.
- LLM (large language model). The engine that reads and writes text.
- Prompt. The instructions given to the model.
- System prompt. The standing instructions: role, facts and rules.
How agents work
- Tool. Something the agent can use: send an email, read a calendar, look up an order.
- Integration. A connection between the agent and one of your systems.
- API. The technical doorway software uses to talk to other software.
- Workflow. A fixed series of steps for a task.
- Trigger. The event that starts the agent, such as a new email.
- Context. The information the agent has in front of it for this task.
- Knowledge base. Your documents and answers the agent can look things up in.
Safety terms
- Hallucination. A confident answer that is wrong or made up.
- Guardrail. A rule that blocks an action or answer.
- Draft mode. The agent prepares; a person approves. See draft mode.
- Human in the loop. A person reviews or approves part of the process.
- Escalation. Handing the conversation to a person.
- Audit log. The record of what the agent did.
Money terms
- Token. The unit models are billed in, roughly a piece of a word.
- Per-resolution pricing. A fee for each conversation the software handles. Ask for the busy-month figure.