Building a Full Mini Project Using Only AI Prompts (Step-by-Step)

Individual prompts produce individual files. To build a real project, you need a sequence of prompts that fit together. In this tutorial we build a small URL shortener end-to-end, using only AI prompts at each stage — and you'll see the exact prompt structure for each phase.

1. Introduction

Most "build with AI" articles either show finished output or skip the messy middle. This one walks through every stage as it happens: planning, scaffolding, building the core, adding tests, polishing, and shipping. The project is small enough to finish in one sitting but real enough to teach you the rhythm of multi-prompt development.

The project: a tiny URL shortener with a Node.js + Express backend, a SQLite database for persistence, and a single HTML page that calls the API.

2. The Concept Explained

Building with AI is not "one giant prompt". It is a sequence of focused prompts, each with a clear scope, each building on the last. The trick is sequencing — you cannot ask for "the next step" until the AI knows what the project looks like so far.

The key discipline is to never lose the chat history. Every prompt in this tutorial assumes the AI has the previous prompts and outputs in context — that is what lets each stage be short and focused.

3. The Problem Without This Technique

The naive approach is to ask the AI to "build me a URL shortener" in one prompt. You get something — but it will be a single file that does everything, with no tests, no error handling, and no separation of concerns.

build me a url shortener in node js with html frontend

4. The Solution

--- PROMPT 1: PLAN ---
I want to build a small URL shortener as a personal project.
Stack: Node.js 20, Express 4, better-sqlite3, plain HTML+JS frontend
(no framework). One repo. One folder.

Propose a file/folder layout, name the npm dependencies needed, and list
the 5–6 features the MVP should ship with. Keep it minimal — no auth,
no analytics.

--- PROMPT 2: SCAFFOLD ---
Great. Generate package.json, the SQLite schema (urls table with: id
INTEGER PK, slug TEXT UNIQUE, target_url TEXT NOT NULL, created_at TEXT
NOT NULL), and the initial Express server entrypoint with health check
only. No routes yet.

--- PROMPT 3: BUILD ---
Add two routes:
- POST /api/shorten { url } → returns { slug, shortUrl }. Generate a
  6-char base62 slug. Validate url is http(s) before saving. On collision,
  retry up to 5 times then 500.
- GET /:slug → 302 redirect to target_url, or 404 if not found.

Use better-sqlite3 prepared statements. Insert and lookup helpers in a
separate db.js module.

--- PROMPT 4: TEST ---
Write Jest tests for the slug generator (length, base62 charset) and the
URL validator. Use the supertest library to test both routes against an
in-memory SQLite database. Cover: happy path, invalid URL, 404, collision.

--- PROMPT 5: POLISH ---
Add a single index.html with a textarea for the URL, a button, and a
result area that shows the short URL with a copy button. Vanilla JS,
fetch() the /api/shorten endpoint. Inline CSS, mobile-friendly, focus
states. Errors from the API shown as a red message.

--- PROMPT 6: SHIP ---
Write a Dockerfile (Node 20 alpine, multi-stage if helpful), a docker-compose.yml
to run locally on port 8080, and a one-paragraph README section describing
how to run the project locally.

5. Step-by-Step Breakdown

1. Always start with a plan prompt. Make the AI write down the layout, dependencies, and feature list. This becomes the "constitution" the rest of the project follows.
2. Scaffold before you build. Project structure + package manifest + initial entrypoint. Then routes, then features. Order matters.
3. Build in independent slices. One feature per prompt. Don't ask for "the rest of the app" — ask for one route or one component.
4. Generate tests after the slice, not as an afterthought. AI is much better at writing tests for code it just wrote than for code you pasted later.
5. Save polish for the end. Error messages, accessibility, UI niceties. Trying to polish while still building causes drift.
6. Wrap with a ship prompt. Dockerfile, README, env vars, deploy notes. Even small projects deserve a fresh dev environment two months later.

Tip: Save the entire chat (or a curated transcript). A complete prompt sequence for one project becomes the starting template for the next one. Six well-thought-out prompts are a portable methodology.

6. Practice Exercises

7. Key Takeaways