Tooling
Lab setup
First, make sure you have completed the initial setup.
If you are part of a course
-
Open Terminal. Run the update command to make sure you have the latest code.
$ mwc update -
Move to this lab's directory.
$ cd ~/Desktop/making_with_code/pilot/labs/lab_tooling
If you are working on your own
-
Move to your MWC directory.
$ cd ~/Desktop/making_with_code -
Get a copy of this lab's materials.
git clone https://git.makingwithcode.org/founders-studio/lab_tooling.git
Welcome to Founders Studio! In this lab, you will set up the tools professional software developers use every day, and use them to make your first artifacts: a pitch for a new product, and a slide deck presenting it. You will work with an AI agent from the very start, but you will also learn to see what the agent is actually doing, so that you stay in charge of your own work.
By the end of this lab, you will be able to:
- Navigate your computer and run programs from the Terminal.
- Explain what a coding harness is, and how an AI agent uses tools to read and write files.
- Use git to inspect changes, commit your work, and look back through its history.
- Compare a commercial harness (Claude Code) with an open one (pi).
Into the Terminal
Take a look at your Desktop. Assuming you completed the setup,
you should see a folder called making_with_code. Today we will get to that folder in a
different way, using the Terminal.
The Terminal is a text interface for your computer. Instead of clicking on folders, you type commands. It can feel slow at first, but it is how developers (and AI agents) do most of their work, because commands can be written down, repeated, and combined.
๐ป Open the Terminal and run pwd ("print working directory").
$ pwd
/Users/yourname
This is your working directory, the folder the Terminal is currently "in."
๐ป Run ls to list what is in this folder. You should see some familiar
names, like Desktop and Documents.
๐ป Move into this lab's folder with cd ("change directory"), then list
its contents.
$ cd ~/Desktop/making_with_code/pilot/labs/lab_tooling
$ ls
README.md pitch.md pyproject.toml questions.md slides.md
๐ป Open the same folder in Finder (Mac) or File Explorer (Windows). Run
open . on a Mac or explorer.exe . on Windows. (. means "the current folder.") Compare
what you see in the two windows. They are two views of the same files.
Starting with an agent
Now we will use an AI agent to help write a pitch for a new product. mwc includes a
built-in chat agent, which runs in your Terminal, inside the current folder.
๐ป From inside lab_tooling, start the agent.
$ mwc chat
๐ Open pitch.md in your editor. It is a template with four sections:
the problem, who has it, the proposed solution, and why now.
With your partner, think of a problem you or people you know actually have: something annoying, slow, or confusing in daily life, at school, or at work. Don't worry about whether the solution is realistic yet.
๐ป Ask the agent to help you fill in pitch.md. Tell it about the problem
in your own words. For example:
We want to write a pitch for an app that helps students coordinate rides to after-school
activities. Read pitch.md and help us fill it in. Ask us questions about anything you
don't know instead of making it up.
As the agent works, watch carefully. You will see it do things like read a file, edit a file, or run a command. These are its tools. A language model on its own can only produce text; the program around it (the harness) turns some of that text into actions on your computer, like reading and writing files.
๐ After the agent finishes, open pitch.md again. What changed? Is
everything in the pitch true, and something you actually mean? Edit anything that isn't.
๐ป Now ask the agent to turn the pitch into a short slide deck in
slides.md. The deck uses Marp, a format where each slide is written in
Markdown, separated by ---.
$ marp slides.md --preview
Saving your work with git
Right now your pitch and slides exist only on your computer, and if you change them, the
old versions are gone. git fixes this. A git repository is a folder whose history is
saved as a series of snapshots called commits. Your lab_tooling folder is already a
repository.
๐ป Ask git what has changed since the last commit.
$ git status
On branch main
Changes not staged for commit:
modified: pitch.md
modified: slides.md
๐ป See exactly what changed with git diff. Lines starting with + were
added; lines starting with - were removed.
$ git diff pitch.md
๐ป Stage your changes and commit them with a message describing what you did.
$ git add pitch.md slides.md
$ git commit -m "Draft pitch and slides for ride-coordination app"
๐ป Send your commits to the server.
$ git push
๐ป Make one more improvement to your pitch (by hand or with the agent), then commit and push again. Now look at the history.
$ git log --oneline
Coding harnesses
The mwc chat agent is built on pi, an open-source
harness. Professional developers often use commercial harnesses such as
Claude Code. Both work the same basic way: a language model,
a set of tools (read, write, edit, run commands), instructions, and a loop that lets the model
keep using tools until the task is done.
๐ป Start Claude Code in your lab folder.
$ claude
๐ป Give Claude Code and pi the same small task, one after the other. For example: "Read pitch.md and suggest three ways to make the problem statement more specific. Don't edit the file."
Compare the two:
- Which tools did each one use?
- When did each one ask for permission?
- How were the answers different?
Deliverables
๐ป Answer these questions in questions.md:
- Describe one thing the agent did while writing your pitch that you could not see in its chat messages, and how you found out about it.
- Find one sentence in your pitch or slides that the agent wrote and you changed. Why did you change it?
- In your own words, what is a coding harness? What does it add to a language model?
- What is one difference you noticed between Claude Code and pi?
- Question: Is the built-in chat command
mwc chat, and which model does it use by default? I assumedmwc chat.- Note: This lab assumes
marpis installed bymwc setup.