What agents really are — Agents & Automation
An agent is a loop, not a chat
A chatbot answers. An agent pursues a goal: it plans, uses tools (search, code, apps, a browser), looks at what happened, and decides the next step — repeating until the job is done or it needs you.
Deep research modes, coding agents, browser-operating assistants and automated workflows are all this same loop wearing different clothes.
📝 Read the full lesson notes
An agent is a loop, not a chat
A chatbot answers. An agent pursues a goal: it plans, uses tools (search, code, apps, a browser), looks at what happened, and decides the next step — repeating until the job is done or it needs you.
Deep research modes, coding agents, browser-operating assistants and automated workflows are all this same loop wearing different clothes.
The autonomy ladder
1. Suggest — AI drafts; you do everything else. (Most chat use.)
2. Act with approval — AI proposes each action; you click OK. (Coding agents in default mode.)
3. Act within limits — AI acts freely inside a sandbox or budget; you review results. (Cloud coding tasks, research.)
4. Fully autonomous — rare, and only for low-stakes, reversible work.
Match the rung to the risk. Drafting tweets? Rung 3 is fine. Sending money? Rung 2, forever.
Part of 🤖 Agents & Automation in the free AI Bootcamp.