Cynobi AI Academy
Course 1 · AI Foundations · Lesson 09

Tools, Connectors & Agents

So far, everything ends as text on your screen — a brilliant brain in a sealed glass box. Today the glass opens: tools are the verbs, connectors are the doors, MCP is the plug, and an agent is a goal plus tools plus a loop.

Free · Course 1 ~5 min video Type · Capability unlock

The one mental model

The model still only reads and writes text — it can't click anything. A tool call is: the model asks, the app does, the model reads what happened. An agent is that, with a goal and permission to loop: plan → act → observe → repeat — while you supervise the junctions.

Key terms

Tool
A single action the model may take — search the web, run code, create an event, send an email. Executed by the app, not the model.
Connector
A door into your stuff — drive, email, calendar, company systems — so tools can work on your real files.
MCP
The shared standard for plugging tools into models — the USB port of AI. One port, any tool, any model.
Agent
A goal + tools + permission to loop (plan, act, observe, repeat) until done or stuck. Deep research mode is one you've likely already met.

The misconception to drop

✕
“Agents are magic autonomous AI — set one loose and forget about it.”
✓
Agents fail in brand-new ways — they loop in circles, misread results, and confidently take wrong actions (Lesson 05 never stops being true). Good agent design keeps a human in the loop: it proposes, you approve — especially anything that sends, spends, or deletes.

Put it to work

1
Turn on the tools you'd actually use — web search and file reading first.
2
When you hand over a goal, write it like a brief — Lesson 06 applies double for agents.
3
Keep approval on for any action that touches the real world.
4
Read the log. Thirty seconds of review beats an hour of cleanup. Trust is earned in steps — small, reversible things first.

Ask the AI Tutor

Pause the video and ask anything from this lesson — the tutor answers from this lesson’s material.

How does a tool call actually work? What's the difference between a chatbot and an agent? What is MCP in one sentence? Which actions should always need my approval?