Chat, give files, delegate, let it act: 4 levels of use

This lesson is still being tested: it improves with learners' feedback. If something is unclear, it is the lesson that needs fixing, not you.

By the end of this lesson, you will be able to

tell apart 4 ways of using an AI: chatting, giving files, delegating to an agent, letting it act;
say, for a given task, the simplest level that is enough;
explain why the risk goes up with each level;
know at what point in the course each level is learned.

Why it matters to you

You hear about “agents” that do the work all by themselves, and about chatbots that supposedly “only chat”. The reality is simpler: there are 4 levels, and most everyday needs are handled by the first two, for free. Knowing how to tell them apart saves you from paying for or installing a tool you don’t need.

Before you start


You need	Your assistant, and a document with nothing personal in it: a user manual, a recipe, the page of a public website
Cost	€0: this lesson stays at levels 1 and 2
Duration	30 minutes

Understand it in 3 minutes

Level	What you do	What the AI does	Example	What can go wrong
1. Chat	You write, you copy the answer	It answers in the conversation	Rephrasing a message, finding ideas	A wrong answer, if you don’t check
2. Give files	You add a document or a photo to the conversation	It reads the document, and can create a new one	Summarizing a 30-page user manual	The document is sent to the provider (lesson 3)
3. Delegate to an agent	You open a folder on your computer to the AI	It reads the files in the folder, then creates and edits files, step by step. In Manual mode (the one used in the course), it asks for your approval before editing a file or running a command (except some read-only commands)	Sorting 300 photos into folders by date	A file edited or deleted by mistake. What you write to the agent is sent to the provider, and what it reads in the folder may be part of it (lesson 3): don’t open a folder that contains what you never share
4. Let it act	You allow the AI to act on your behalf, sometimes without you watching	It sends, publishes, books, replies, at a set time or on its own	Preparing a summary of your emails every morning	An action that cannot be undone: an email sent, a purchase, a deletion

An agent is an AI that does not just answer: it takes actions, one by one, to reach a goal. Claude Code (from Anthropic) and Codex (from OpenAI) are agents that work in your files.

Where the free plans stand, as of 24/10/2026:

levels 1 and 2: Claude’s free plan lets you chat, search the web, and create files (claude.com/pricing). ChatGPT Free lets you add files, with limits (chatgpt.com/pricing). For free Claude, the pricing page does not say: to be checked in the app;
level 3: Claude Code is included starting with the Pro plan (claude.com/pricing). Codex is included in the ChatGPT plans, with limited access on the free plan, for small programming tasks (learn.chatgpt.com/docs/pricing, chatgpt.com/pricing);
level 4: depends on the tools. For example, Hermes Agent is free, open source software, but the AI it uses may be paid (hermes-agent.nousresearch.com).

The course’s rule: move up a level only when the previous level is no longer enough. And at level 4, nothing irreversible without your approval.

Do it together

1Step 1: Sort some tasks

For each task, choose the simplest level that is enough. The answers are in the “Check your work” section.

Writing an apology message for being late.
Understanding an electricity bill that you have as a PDF.
Renaming and sorting 300 vacation photos by date.
Finding 10 gift ideas for a birthday.
Sending a summary of your calendar to your family every Monday, without thinking about it.
Turning an expenses table into a chart.

2Step 2: Try level 2

In your assistant, find the ”+” button near the text box, then the entry to add files (“Add files or photos” on Claude, “Add photos & files” on ChatGPT).
Add your document with nothing personal in it.
Copy this request:

Summarize this document in 5 points, for someone who has never read it.
Then tell me what is missing or unclear in this document.

Check one point of the summary by rereading the passage of the document that covers it: did the AI understand it correctly?

Your turn

List 5 real tasks from your week or month that could benefit from an AI.
For each one, write down the simplest level that is enough, and why.
Do one of them right now, at level 1 or 2.
Write in your ID sheet the level that covers most of your current needs.

Check your work

See the answer for step 1

Answers for step 1:

Level 1: chatting is enough.
Level 2: you need to give it the file (without your unnecessary personal information, or knowing your privacy setting).
Level 3: 300 files to rename is the job of an agent. By hand in a conversation, it would be endless.
Level 1.
Level 4: a repeated action, without you. This is module 8, with its safeguards.
Level 2: you give the table, it gives you back a chart or explains how to make one.

Your exercise worked if each of your 5 tasks has a level and a reason, and if you did one for real.

Common pitfalls:

looking for an agent for a task that a simple conversation handles;
giving a personal document “just to see”, without having looked at your privacy setting;
believing an agent works unsupervised: at the start, you approve each step.

Keep it and pick it up later

Keep your list of 5 tasks: it will help you choose your path, and know when module 4 (the agent) will be worth it for you.

Going further

optional

The page Choosing your tool in this course compares Claude, ChatGPT, Claude Code, Codex, and Hermes, with dated information.

New words

5 words explained

Word	What it means
Agent	An AI that chains actions (reading, creating, editing, sending) to reach a goal
Working folder	The folder on your computer where the agent works. In Manual mode, to edit elsewhere, it asks for your approval, unless a permission you already gave covers that action. To read elsewhere, it usually asks too, but some read-only commands (listing files, displaying one) can read elsewhere without asking. A command you accept can touch other folders: read each request
Permission	The approval you give, or refuse, before an action by the agent
Automation	A task that starts by itself, at a set time or after an event
Irreversible	An action that cannot be undone: sending, a payment, a permanent deletion

Still to verify: adding files on the free plans of Claude and ChatGPT (exact limits), the location of the add-file button.