From idea to app — building software with AI.
Class 1 · 12 weeks · 3 blocks
You get a real codebase on day one. You ask it questions, you find your way around, you change one thing and see it work.
You add features, first to our app, then to your own. AI writes most of the code. You decide what to build and whether it's right.
Something deployed, with a URL, that other people can click. And you can explain every part of it.
Lists, loops, functions, classes — you'll meet all of them. Not to memorise them, but to read what the AI wrote, catch it when it's wrong, and tell it what you actually want.
Block 2: build one of three cases, alone. Block 3: your team extends the best of your solo projects. Details when we get there.
| 1 | THEORY | Introduction and workflow design — today. |
| 2 | PRACTICAL | Going through the apps + asking questions — open the codebase, interrogate it. |
| 3 | THEORY | How to build a feature, and git — watch one get built and saved. |
| 4 | PRACTICAL | Build it yourself, and git — your turn. |
Live demo at a URL, then a code walkthrough where every member explains code they didn't write. One grade per group, adjusted per person from the walkthrough.
30% — questions about your Block 2 solo project: what you built, why, where things live, what would break.
30% — general: read AI-generated code and judge it. Find the bug, predict the output, spot the missing case.
Your solo project is never submitted — it's examined. Build it well enough that you can answer for it with no laptop in front of you.
The rest of today: learn what a workflow is, then design one yourselves. The apps come next week — in the first practical.
Most people who fail at building software don't fail at code — they fail at describing the process. AI can write the code. It cannot decide your workflow.
What starts it? Someone pays. A lead calls. A form arrives.
What happens, in what order? One box, one action.
Where does it branch? Every branch needs an answer for both sides.
What does each step need to know — and where does that come from?
Four friends go on a weekend trip. They pay for things at different times, in different amounts — and some things are only for some of them. Design the workflow that ends with everyone square.
Groups of 3–4, paper or whiteboard. Boxes for steps, diamonds for decisions, arrows for order. For every step, write the information it needs — and where that information comes from.
No code, no app names. One clear trigger, one clear end state ("everyone is square"). Every decision has both branches drawn. If a step needs data nobody wrote down earlier, your workflow has a hole.
Trip, person, expense, share. The things your steps talk about become the tables — db.py.
"Add an expense", "see who owes whom". Where a person acts, the app needs a screen — app.py.
Who's in this split? How much each? The branches and calculations become the rules — logic.py.
A bad workflow makes a bad app, no matter how good the AI. A precise one is already half the prompt. Next class, you'll open an app that was built from exactly this workflow.
Groups of 3–4 form next week, in the room. Think about who you want.