The point of doing it twice

You are about to build the identical thing, from the identical request, in a completely
different tool.

This is not to decide which is better. It is so that you can see which parts of what you
learned yesterday were about agents and which parts were about one product’s screen
layout
.

Set up a separate folder

mkdir recall-antigravity
cd recall-antigravity
git init

Keep it separate from your first attempt. You will compare them in the next lesson.

What is different before you even start

Claude Code lives in your terminal. Antigravity is a code editor, so the agent works beside
your files, and you can see the project, the changes and the agent’s progress in one window.

It leans more on written plans you approve before work begins, and it can open a browser to
look at what it built.

Different surface. Same four beats underneath. Watch for them.

Use the exact same request

Paste the same request from Lesson 3.4. Word for word. Changing it would ruin the
comparison.

What to watch for

The same four questions, plus three about the difference:

  • Where do you see the plan? Is it a written document you approve, or text going past?
  • How do you approve actions? Is it per action, or once for a batch?
  • Can you tell what it read? Or does it read quietly and only show you results?

Record this

Fill in the same table as before.

Question Your answer
How many files did it create?
Did it use any library you did not ask for?
Did it start the server itself to check?
How many times did you approve something?
Did you read every command before approving?
Did anything happen that you did not expect?

Commit

git add -A
git commit -m "Step 1: empty project serving Recall (Antigravity)"