Lesson 3 — Working with an AI Assistant
From a chat to a collaborator
Lesson 1 was about what an AI assistant is. This lesson is about how to work with one on something real — a document, a website, a small program. Modern AI assistants can do more than answer questions: they can read your files, make changes to them, and explain what they find. That makes them powerful, and it makes the way you direct them matter. The habits below apply whether you are using Claude, ChatGPT, or any other assistant.
1. Give the AI context first
An AI does not automatically know what you know. Before asking for a change, say what you are trying to do and why. The goal is not a longer message — it is a message the AI could act on without having to guess.
Try it yourself — which of these requests could an AI act on, and which would leave it guessing?
This activity needs JavaScript enabled. The lesson still covers everything.
2. Scope one task at a time
A request like "redesign the whole site and add three features" is too big to do well or to review. Break work into single, checkable steps. Finish one, look at the result, then start the next. Small steps are easier to verify and easy to undo if something is wrong. This habit becomes the backbone of the build process in Lesson 5.
3. Pause before anything risky
Reading files, drafting text, and editing a page are low-risk — easy to review and undo. Other actions are not: deleting files, changing a database, publishing something public. For anything hard to undo, slow down, confirm it is really what you want, and ask for help if you are unsure. The old carpenter's rule fits well: measure twice, cut once.
Sort these actions — which can you let the AI run, and which deserve a pause?
This activity needs JavaScript enabled. The lesson still covers everything.
4. Read what the AI did — do not just accept it
When an AI makes a change, it will describe what it did. That description is what the AI intended; the actual change is in the files. Open the file, or look at the result, and confirm it matches. This habit catches the confident-mistake problem from Lesson 1 before it causes any damage.
5. When something is wrong, say what you see
If a change is not right, describe the actual symptom — what you expected, what happened instead, any error message. That is far more useful than "it's broken," and it lets the AI find the real cause instead of guessing.
Quick check
This activity needs JavaScript enabled.