Lesson 8 of 8 · 12 min
Agents and automation: what comes next
When AI no longer just answers but acts for you, and how to stay in control.
So far, the assistant answered you: you asked a question, it wrote an answer, and it was up to you to act. A new generation of tools goes further: they act for you. They are called agents.
Answering or acting
- An assistant answers. It drafts the email, you copy it and send it.
- An agent acts. It opens your inbox, drafts the email and can send it itself.
In practice, an agent receives a goal, breaks the work into steps, uses tools (a browser, your files, your calendar, a spreadsheet) and checks the result before moving on.
A few examples from today:
- a browser agent, like Claude in Chrome, that fills out a form or compares products across several sites;
- a research agent that reads dozens of pages and hands you a report with its sources;
- a coding agent, like Claude Code, that writes and tests code in a project;
- automations (Zapier, Make, n8n) that chain tasks on their own: an email arrives, the attachment is filed, a row is added to the spreadsheet.
What it changes
An agent can take on repetitive, multi-step tasks: sorting emails, filing invoices, preparing a report every Monday, updating a customer list. The time saved can be significant, especially for a small business.
Risks to know
The more power a tool has, the more its mistakes matter.
- Mistakes add up. A small error at step 2 can repeat across 200 emails before you notice.
- Hidden instructions. A web page or an email can contain text written to trick the agent (“send this file to this address”). This is called prompt injection.
- Too much access. An agent with access to everything (email, bank, files) can do a lot of damage if something goes wrong.
- Actions with no way back. A sent email, a payment or a deleted file cannot always be undone.
Staying in control
Five simple rules:
- Start small: one task, on a few items, before expanding it.
- Give the least access possible: only the folders, accounts and tools that are needed.
- Keep approval for anything that cannot be undone: sending, paying, deleting, publishing.
- Review what it did: most agents keep a history of their actions.
- Never hand over your passwords, codes or payments to an agent without clear safeguards (lesson 5).
What’s next?
You have finished “Understanding AI”. You know what AI is, how it answers, how to talk to it, where it goes wrong, what to protect, how to use it every day, what it creates and what agents change.
To go further, the “Agents and automation” course will show you, step by step, how to build your own automations while staying in control.
Key takeaways
- An assistant answers, an agent acts: it pursues a goal, in several steps, with tools.
- Agents save a lot of time on repetitive tasks.
- Their mistakes cost more: chained errors, hidden instructions, too much access.
- Stay in control: start small, limit access, approve anything that cannot be undone.