Build the One-Screen Version of Your App First
You do not need a database, user accounts, or a cloud subscription to see if your app idea works. Here is how to use an AI assistant to build a local, one-page prototype in an hour, and test it on your friends before you spend a weekend building the wrong thing.

The neighbourhood tool library has three drills, a hedge trimmer, and a WhatsApp group full of messages that say, “I think Sam still has it.” The natural instinct of the enthusiastic tinkerer is to immediately build a booking app. You could spend an entire weekend setting up user accounts, email notifications, and a cheery little map.
Or, you could spend a single evening finding out whether your neighbours can actually answer one simple question: Is the drill free on Saturday?
When you have an idea for a digital tool—whether it is a roster for the school carpool, a holiday booking system for your small business, or a shared flat chore wheel—the urge to build a finished product is overwhelming. But building the entire system before you know if people understand the core concept is a recipe for wasted time. According to the UK Government Digital Service’s (GDS) manual on agile delivery, you do not have to prototype a user’s entire journey. The advice is to make a prototype sufficiently detailed to investigate the uncertain parts of an idea, with production engineering coming later.
This recipe works for any idea where you can point to one person, one decision, and one screen. You are going to use an AI coding tool to generate a single-file, interactive mockup that runs entirely on your own computer, and then you are going to run a brief usability test with a friend to see if the idea actually survives contact with a real human.
The Finish Line
A single-file HTML web application running on your local computer, a complete testing script, and notes from a usability session to tell you whether your idea is worth pursuing.
- Time: 60 minutes (20 minutes to generate and run the code, 40 minutes to prep and run the test).
- Cost: Free.
- Prerequisites: A standard conversational AI tool (like ChatGPT, Claude, or Copilot), a basic text editor (Notepad for Windows or TextEdit for Mac), a web browser (like Chrome, Safari, or Edge), and one willing friend or colleague to test your creation.
1. Identify your riskiest assumption
A crucial part of this prototyping phase—often called an "alpha" phase—is identifying your riskiest assumptions and testing them. What is the one thing that, if it fails, ruins the whole project?
For the neighbourhood tool library, the riskiest assumption is not the login screen. It is whether a neighbour who needs a tool can easily tell if it is free on a chosen date, without messaging anyone. That is the only job this prototype needs to do. Multiple days, refunds, late-return penalties, and a beautiful sign-up screen can wait.
On a scrap of paper, write down the single task your user must achieve. For this guide, our task is: A user selects a calendar date and successfully reserves an available tool.
2. Generate the single-page prototype
AI coding assistants are incredibly eager to overcomplicate things. If you simply ask for "a booking app," they will often try to give you Python scripts, database schemas, and deployment instructions that require a computer science degree to decipher.
You must place strict boundaries on the generation. You want a plain HTML file that uses basic JavaScript to mimic interactivity. It will not remember data if you refresh the page, and that is perfectly fine for a prototype.
Copy and paste this exact generation brief into your AI chat tool, adjusting the details for your own idea if necessary:
Copyable Prompt:
Make a one-page tool library demo I can run locally in my own web browser. Use plain HTML, CSS and JavaScript, all contained within a single file. Do not use external libraries, databases, or backends.
Seed the interface with three fictional tools (a drill, a sander, a hedge trimmer) and one existing dummy booking for the drill on 14 February 2027.
A user must be able to select a calendar date and see which tools are available. They can click to reserve an available tool for that day in the current session. Reject a second reservation for the same tool and date. Add a simple "Reset" button at the bottom that clears their current session selections. Use labelled text for availability (e.g., "Available", "Booked"), not colour alone.
No login screens, no real names, no external services. Output the entire code block so I can copy it into a single text file.
3. Save and run your app locally
The AI will reply with a large block of code. You do not need to understand the code to run it, but you do need to save it correctly.
- Copy the code block generated by the AI.
- Open your computer's built-in text editor. On Windows, use Notepad. On a Mac, use TextEdit. (Crucial step for Mac users: go to the menu bar, click Format, and select Make Plain Text. If you skip this, your Mac will try to save a rich-text document, which a web browser cannot read as an app).
- Paste the code into the empty document.
- Save the file to your Desktop. Name it
library-prototype.html. You must type the.htmlextension yourself, and ensure your text editor does not secretly append.txtto the end. - Close your text editor. Find
library-prototype.htmlon your Desktop, right-click it, select Open With, and choose your web browser (e.g., Chrome or Safari).
Illustrative Output: Your browser will open what looks like a real, functioning website. You should see a date picker, a list of three tools, and a reset button. Clicking the drill on 14 February 2027 should show it as already booked.
4. Run a moderated usability test
Now that you have a functioning screen, invite a friend or neighbour to test it. This is where you learn if your design actually makes sense.
The GDS guide to moderated usability testing outlines how to do this effectively. Moderated usability testing involves watching participants try to complete specific tasks while asking them to "think aloud."
Before you begin, sit your tester down and run through a brief, reassuring introduction. The GDS manual advises you to explicitly let them know that you’re testing the service, not them. Reassure them that they are not being judged or assessed. If they get stuck, it is the app's fault, not theirs.
Do not give them a guided tour of the buttons. Instead, give them a realistic, goal-oriented task.
Read them this instruction:
"You need to borrow a drill on the 14th of February 2027. See if you can get it. Please tell me what you are thinking as you look at the screen."
Now, try to stay completely quiet. Mostly just watch and listen. Note where they hesitate. If they click a date but do not realise they have to click a separate "Reserve" button, write that down. If they get completely stuck, the GDS suggests you can help them recover so you can continue learning, but only after observing the initial failure.
Checks that show it worked
- The local file opens: The
.htmlfile opens cleanly in your browser without showing raw code text. - The state changes: Your tester can click a tool, and the text actively updates to say "Reserved" or "Booked."
- The reset works: Clicking the Reset button clears the tester's actions, returning the interface to its original dummy state, proving the local JavaScript is functioning.
Common failures and recovery
The browser just shows a wall of raw text.
You saved the file as rich text (RTF) or a plain text document (TXT) instead of HTML. Go back to your text editor. If on a Mac, ensure you selected Format > Make Plain Text before pasting. When saving, manually type .html at the end of the file name and confirm any popup warnings asking if you want to change the extension.
The AI gave me multiple files or instructions to "install Node.js".
The AI ignored your constraints and tried to build a production app. Reply to the AI with: "Stop. I do not want a backend, Node, Python, or multiple files. Rewrite this as a single HTML file with embedded CSS and vanilla JavaScript that I can open directly in Chrome by double-clicking it."
The tester gets confused and apologises.
This is a testing failure, not a code failure. If your tester says "Sorry, I'm just bad with computers," immediately stop and use the GDS recovery tactic: remind them they are helping you find flaws in your design. Ask open-ended questions like, "What did you expect to happen when you clicked that?"
Saving, exporting, and the art of throwing it away
If your tester successfully navigated the calendar and understood the tool availability without your help, you have answered your riskiest assumption. Your interface makes sense.
Now comes the most important step: do not try to turn this file into a real app.
The GDS approach treats prototype code and early ideas as disposable. This file is a sketch of an interaction. It has no shared database or production access controls by design; that does not make generated code safe for real data.
You do not need to maintain this code. You can safely delete it. What you keep are the notes you took while watching your friend navigate the screen. If they proved the concept works, you now have the confidence—and the precise screen requirements—to move on to a beta phase, where you can safely invest the time into building the real, database-backed tool library knowing the core interaction is sound.
AI disclosure: AI systems assisted with research and writing. Gemini independently peer-reviewed and approved this article before publication. Images, where present, are illustrative AI-generated photographs.