AI Agent Glossary: Plain-English Terms

Ever read an AI tool's website and felt like it was written in another language? You're not alone. Here are the words you'll see most, in plain English. Where a tool uses a term in its own way, we say which tool (we checked each page on Oct 28, 2026 or Oct 27, 2026).

The basics:

• AI agent: AI that does a task for you, step by step, using tools like your email or a spreadsheet, instead of just answering a question.
• Chat assistant (or chatbot): AI you talk to in a chat window. It answers and writes, but usually doesn't go off and do the work in your apps.
• Bot: xAI's name for an AI teammate in Grok Bot. xAI says Bots have their own computer, so they can work inside your apps and tools.
• Model: the AI "brain" underneath, made by companies like OpenAI, Anthropic, Google, or xAI. Many agent tools let you pick which one to use.
• Prompt (or instructions): what you tell the agent to do. Clear, specific instructions get better results.

How agents get work done:

• Workflow: a set of steps that runs in order, like "new form entry, look up the customer, draft an email."
• Trigger: the thing that starts a workflow, like a new email, a schedule, or a form entry.
• Tool (or integration): an app the agent is allowed to use, like Gmail, Slack, or HubSpot.
• MCP (Model Context Protocol): an open standard for connecting AI apps to other systems. The official MCP docs compare it to a USB-C port for AI apps.
• API: a way for software to talk to other software. You'll see it on developer pages. You usually don't need it for no-code tools.
• Webhook: a simple message one app sends another when something happens, often used as a trigger.
• Routine: a task an agent repeats on its own. xAI says you can show a Bot a workflow once and it saves it as a routine.
• Skill: a saved, reusable way of doing a task. Lindy and Gumloop both use this word.

Keeping you in charge:

• Human in the loop (or approval step): a point where the agent stops and waits for a person to say yes. Zapier, Make, Lindy, and Microsoft Copilot Studio all describe these.
• Handoff: when an AI agent passes a conversation to a person. Botpress calls it a hot handoff when the full history goes along with it.
• Run history (or logs): a record of what the agent did, step by step. Handy when something looks off.
• Memory (or context): what the agent remembers about you and your work between tasks.

Where your info lives:

• Knowledge base: documents you give an agent so it can answer from your own info. Dify calls the prep work for this a knowledge pipeline.
• Self-hosted: you run the software on your own server instead of the company's cloud. Dify's Community Edition and n8n both offer this.
• SSO (single sign-on): one company login for many apps.
• DLP (data loss prevention): rules that help keep private info from leaking. xAI says Enterprise admins can set DLP for Grok Bot.

What it costs:

• Credits, tasks, and conversations: the units tools charge by. Lindy and Gumloop use credits, Zapier counts tasks, Botpress counts conversations, and Dify uses message credits. They're not the same size, so always check what one unit buys.
• Bring your own key: using your own account with an AI model company, so you pay them directly for the AI part. Gumloop and Dify both mention this.

That's the list for now. Pick one tool, read its page with this glossary next to you, and it'll make a lot more sense.