Your new intern

Today’s finish line

Before we leave, you will have pointed an AI agent at a real folder, asked it to understand the mess, changed its proposed plan, and watched it carry out the work.

No previous terminal experience is assumed.

Morning — first contact

10:00 · See the whole thing

We begin with a short live demonstration: a chaotic course folder goes in; a clean, documented project comes out. Do not worry about how each step works yet. First, see the shape of the possibility.

10:30 · Find your feet

Open a terminal. Move into a folder. Start Claude Code or Codex. Ask:

What files can you see here? Do not change anything.

That small exchange contains most of the technical foundation we need.

11:00 · Tiny, harmless jobs

Ask the agent to create a Markdown file, inspect it, revise it, and remove it. Try speaking your instructions instead of typing them.

Useful failure is welcome: if something breaks, ask the agent what happened and what it wants to try next.

Afternoon — decrap a folder

You will receive the same deliberately awful folder as everyone else: duplicates, mystery names, old exams, stray images, and several competing versions of “final”.

Work through this loop:

1. Inspect: “Describe what is here. Do not change anything.”
2. Plan: “Propose a sensible structure and explain your choices.”
3. Disagree: Change at least one part of the plan.
4. Act: Let the agent carry it out.
5. Check: Open the result and inspect a few files.
6. Document: Ask for a short README.md explaining the new structure.

The two-minute terminal kit

You only need a handful of moves today:

Tab completes names. The up arrow recalls your previous command. When unsure, ask the agent.

Before you leave

• Keep the cleaned folder.
• Keep the README.md.
• Write down one instruction that worked surprisingly well.
• Write down one thing you did not trust.

Tomorrow we take the same inspect–plan–act–check loop into research.