Build With AI

Level 4 · Build With AI

1. Concept: Describe it in plain words

   You explain the tool the way you'd explain a job to a new hire, and the helper turns your words into code, one small step at a time.

   1. Term: AI coding helper

      An AI that reads the files in your project, writes changes and can run commands, usually after asking you.

      1. In plain words: A fast new hire

         Think of a fast, eager new hire who has read a mountain of code but has never seen your business. You say what you want, and it does the typing. Techies call it a coding agent.

      2. How it works: Read, plan, edit, test

         You type a request in a chat panel. The helper reads the files in your folder, makes a plan, edits the files and may ask to run a command to try them. You look over each change, then keep it or undo it.

      3. What it costs: Free to learn on

         GitHub, Google and Mistral all have free plans that work on the class laptops, with usage limits. Plans that raise those limits run about $10 to $20 a month (checked Oct 23, 2026).

         See also Free plans, Paid helpers

      4. Watch out: Confident, not always right

         It sounds just as sure when it's wrong. It can misread you, or change more than you asked for. You're the boss of every change.

         See also Test with numbers you know, Diff

   2. Term: A good first project

      A small chore with clear rules and an answer you can check, like pricing a job.

      1. In plain words: Fits on a napkin

         Pick a chore you already do by hand that follows clear rules, like "$5 per 1,000 square feet, $40 minimum." If you can explain it on a napkin, a beginner can build it in an afternoon.

      2. Try it: Three starters on the kit

         The class laptops will come with three starters: a price-quote calculator web page, a script that sums up today's appointments from a spreadsheet file, and an inventory checker that lists what to reorder. All three use made-up sample data.

      3. Watch out: Not on day one

         Skip anything that takes payments, stores customer passwords or goes on the internet for strangers to use. Those need security know-how, so save them for later, with help.

   3. Term: One small step at a time

      Ask for one change, test it, save it, then ask for the next.

      1. In plain words: One wall at a time

         Build it the way you'd frame a shed: one wall, check that it's square, then the next. A small ask is easy to check. A giant ask makes a giant mess to untangle.

      2. How it works: Start small, then grow

         Real tools grow this way. Alex's app that turns handwritten open-house sign-in sheets into contacts for a local real estate agent began as a single screen: snap a photo, check it, send it. Extras, like a password screen, came later, one at a time.

      3. Try it: A good first ask

         Try: "Make a one-page price calculator with a box for lawn size in square feet and a Calculate button. Charge $5 per 1,000 square feet, with a $40 minimum." Test it before you ask for anything else.

         See also Test with numbers you know

      4. Watch out: When it goes in circles

         If the helper keeps "fixing" the same problem and making it worse, stop. Go back to your last save point and ask a smaller, clearer question.

         See also Git save points

   4. Term: Project notes file

      A short text file in your folder that tells any AI helper your rules, so you don't repeat them every time.

      1. In plain words: House rules

         It's the note you'd leave a new hire: what the tool is for, who uses it and what not to touch. Many helpers look for a file named AGENTS.md.

      2. How it works: Read at the start

         When you start a chat, the helper reads the notes file along with your request. Copilot in VS Code, OpenAI's Codex and Claude Code all read AGENTS.md (checked Oct 23, 2026), so your rules come along when you switch helpers.

         See also Free plans, Paid helpers

      3. Try it: Add one rule

         Open the starter project's AGENTS.md and add one line about your business, like "Use friendly, plain words on the page." Then ask for a change and see whether the helper follows it.

2. Concept: Your workshop: folder, editor, terminal

   Three things on the laptop do the work: a folder that holds the project, an editor to see and change it, and a terminal to run it.

   1. Term: Project folder

      One folder holds every file for one tool, and it belongs to you, not to the AI company.

      1. In plain words: The folder is the tool

         Your tool is a handful of plain files in one folder. Copy the folder and you've copied the tool. You don't need an AI account to keep it or run it.

      2. How it works: The helper's fence

         Open a folder in the editor and the helper works inside it, reading and editing those files. That's a good fence, not a perfect one, because commands it runs can reach further.

         See also Terminal

      3. Watch out: Keep secrets out

         Files in the folder can be sent to the AI company while the helper works. Keep customer lists, passwords and tax files somewhere else, and build with made-up sample data.

         See also Keep private info out

   2. Term: Code editor

      A program for reading and changing code; the class laptops use VS Code.

      1. In plain words: A word processor for code

         Files on the left, the open file in the middle, the AI chat on the side. VS Code is free, comes from Microsoft and runs on Windows, Mac and Linux.

      2. How it works: One screen, many helpers

         VS Code's chat panel can use GitHub Copilot or a model running on your own laptop. Terminal helpers like Google's Antigravity run in its terminal panel. Learn one screen, then swap the helper.

         See also Free plans, Local models

      3. Try it: Find the three parts

         Open the starter folder in VS Code. Find the file list and the chat panel, then choose Terminal, then New Terminal, from the menu at the top.

   3. Term: Terminal

      A window where you type short commands instead of clicking buttons.

      1. In plain words: Typing instead of clicking

         The terminal is a text window for giving the computer orders, like "run this" or "show me the files." It looks old-fashioned because it is. It's also the quickest way to do a lot of jobs.

      2. How it works: Command, Enter, answer

         Type a command, press Enter, and the computer prints its answer. ls lists the files in the folder you're in, and python3 today.py runs a Python script called today.py.

      3. Watch out: Read before you allow

         Helpers ask before running risky commands, so read each one. If it deletes files (rm), installs software or touches passwords and you don't know why, say no and ask what it's for. Skip Allow all while you're learning.

3. Concept: Check it, then save it

   Test every change with cases you can work out yourself, look at exactly what changed, then save a checkpoint you can go back to.

   1. Term: Test with numbers you know

      Try inputs where you already know the right answer.

      1. In plain words: Do the math in your head

         If 10,000 square feet at $5 per 1,000 should come to $50, type 10,000 and look. If the page says $500, you caught the bug before a customer did.

      2. Try it: Three test cases

         Before each new feature, write down three tests: a normal job, a tiny job that should hit the minimum charge, and an empty box. Run all three after every change.

      3. Watch out: "It works!" is a claim

         Helpers often announce success without trying the thing. Run it yourself, and ask the helper to explain any number you can't check.

         See also Diff

   2. Term: Diff

      The before-and-after view of a change, line by line.

      1. In plain words: Track changes for code

         Like track changes in Word: added lines show in green, removed lines in red. It shows what the helper actually did, not what it says it did.

      2. How it works: Keep or Undo

         When the helper edits a file in VS Code, the edit is saved right away but marked as pending, with Keep and Undo buttons. The Source Control panel shows every change since your last commit.

         See also Git save points

      3. Try it: Spot the surprise

         After your next change, open the diff and find one line you didn't expect. Ask the helper: "Why did you change this line?"

   3. Term: Git save points

      Git saves snapshots of your whole project so you can go back to any of them.

      1. In plain words: Save points

         A commit is a save point with a short note, like "added minimum charge." If the next change breaks things, you go back to it. Git is free and lives in your project folder.

      2. How it works: A hidden history

         Git keeps its history in a hidden .git folder inside your project. Each commit records what changed since the one before. In VS Code, you type a short note and click Commit in the Source Control panel.

      3. Try it: Break it on purpose

         Commit a working version. Ask the helper for a change, then right-click the changed file in Source Control and choose Discard Changes. Your working version comes right back.

         See also One small step at a time

      4. Watch out: Not a backup

         Git lives on the same laptop, so if the laptop dies, the history goes with it. Copy the folder to a USB stick now and then, or push it to a private repository on GitHub.

4. Concept: Pick a helper, keep your project

   Several companies make coding helpers and some AI runs right on your laptop; no AI company pays Iowa AI, and your folder works with any of them.

   1. Term: Free plans

      Three free cloud helpers work on the class laptops, each with a usage limit.

      1. In plain words: Free, with a meter

         GitHub Copilot Free, Google's Antigravity CLI and Mistral's Vibe all have free plans that run on Linux. Each one stops or slows down after a set amount of use.

      2. What it costs: $0, with limits

         Copilot Free gives 2,000 code suggestions a month plus a chat allowance GitHub doesn't put a number on. Antigravity's free plan lists "basic weekly rate limits." Checked Oct 23, 2026, and these change often.

      3. Watch out: Free plans move

         In June 2026, Google ended free sign-in for its Gemini CLI and pointed people to Antigravity. Learn a second helper, and keep your work in your own folder.

         See also Project folder

   2. Term: Paid helpers

      About $10 to $20 a month buys more use and stronger models, from tools like Claude Code and OpenAI's Codex.

      1. In plain words: More room to work

         Paid plans give you more requests and stronger models. That matters once you build every week, not on your first day.

      2. What it costs: $10 to $20 a month

         Claude Code comes with Claude Pro: $20 a month, or $17 a month paid yearly. OpenAI's Codex terminal tool comes with ChatGPT Plus at $20. Copilot Pro is $10. Checked Oct 23, 2026.

      3. Watch out: Still metered

         Paid plans have limits too, often on a five-hour clock plus a weekly cap. Use the free plan for a couple of weeks and see whether you actually hit the wall.

         See also Free plans

   3. Term: Local models

      Yes: with Ollama, a small open model runs right on your laptop, so nothing leaves the machine.

      See also Keep private info out

      1. In plain words: The AI stays home

         Ollama is a free program that runs open AI models on your own computer. Your code and questions never leave the laptop, and it works with the Wi-Fi off.

      2. How it works: Same chat panel

         Install Ollama, download a small model such as qwen3.5:4b, and add the Ollama extension to VS Code. The model shows up in the same chat panel, with no account needed.

         See also Code editor

      3. What it costs: Free, paid in patience

         The software and the models are free. On these laptops, with no graphics card, expect a minute or more per coding answer (our estimate, so we time each laptop before class).

      4. Watch out: Small models, small jobs

         Laptop-sized models trail the big cloud ones by a long way. Use them to explain code or make a small fix, not to build the whole tool.

         See also One small step at a time

   4. Term: Keep private info out

      Cloud helpers send your code and questions to the AI company, so keep secrets and customer details out of the chat.

      1. In plain words: A postcard, not a letter

         Anything you type, and any file the helper reads, goes to the company's computers, and some companies let staff review chats. Treat it like a postcard, not a sealed letter.

      2. Try it: Flip the training switch

         Since April 24, 2026, Copilot's free plan may use your code and chats to train AI models unless you opt out. On GitHub, click your picture, choose Copilot settings, and set "Allow GitHub to use my data for AI model training" to Disabled.

      3. Watch out: Never paste these

         Never paste passwords, API keys, customer names and phone numbers, Social Security numbers or bank details. Build with made-up data. The finished tool can use real data because it runs on your laptop, and the AI chat never needs to see it.

         See also Project folder