Part II: Building the first version
11. Choose your tools and get set up
Set up your accounts, prepare the build brief, and get a starter running in your browser.
Have your problem sentence, first useful flow, and rough screen sketch nearby. These are the instructions for what you're about to build. Keeping them together helps you give the AI consistent context when you start making changes.
The aim of setup is modest: a project you can open, a way to ask the AI to change it, and a preview where you can see the result. Get that working before adding your product's behaviour.
We'll use the browser versions of these tools. The AI runs the development tools on a cloud computer, so you can give instructions and review the result without installing Node.js or Bun, or opening a terminal on your own computer.
Create your Prisma account
We'll use Prisma to create and host the application, with Prisma Postgres for its database and Prisma Compute to run it. Its Free plan includes both, with usage limits and no credit card required.
- Open Prisma Console. Create an account if you don't have one.
- Sign in with your email and password, Google, or GitHub. Choose whichever you prefer.
- Prisma creates a default workspace for you. A workspace groups your projects and their billing. Use this personal workspace for the handbook and keep it on the Free plan.
You should now be able to open your workspace in the Console. You can leave it empty for now; we'll connect the AI before creating the product's database. The Prisma account setup guide has more detail if you get stuck.
Set up your AI account
Choose the tool above and follow its steps. You only need one AI subscription. If you already pay for it, check what your existing account includes before buying anything else.
Choose ChatGPT or Claude above to see these steps.
Prices and included usage can change. Check the amount and billing period at checkout. Start with monthly billing if you're trying a paid plan, and keep extra paid usage off while you learn how much you need.
The cloud computer also needs permission to reach the services used to build and deploy your product. In ChatGPT Work, this depends on your Work network settings and any workspace rules. Claude's Default environment uses a list of trusted destinations. If a build or deployment reports blocked network access, ask the AI which destination it needs and why before changing a setting. A connected Prisma account doesn't by itself establish that this separate network access is available.
Put the brief in one place
Your notes now contain the person and problem, your proposed solution, and the steps your first version needs to support. Bring those together into a short build brief. Include the features you're leaving out and the check you'll use to tell whether the task works. Add your sketch if the tool supports it, or describe the screen in words.
For our hypothetical example, the brief could be:
We're building a way for a freelance designer to ask a client
for a decision on a design draft.
The designer creates a request with a draft URL and a question.
The client opens the request and submits a written response.
The designer can read the saved response later.
Start with the request page using clearly labelled sample content.
We'll connect the complete flow to saved data afterwards.
Keep the layout simple and usable on a phone. Explain what the
client is responding to and confirm when a response is saved.
Leave out payments, automated reminders, file uploads, analytics,
and team management for this version.Change the brief to match your own product. The excluded features help the AI understand where to stop. Keep a copy with the project so you can refer to the same decisions later.
The prompts in the following chapters continue our designer example. Before using one, replace the people, information, and actions with the ones in your brief. You can ask the AI to adapt a prompt to your brief and show it to you first. Read it before asking the AI to build, so it doesn't add parts of our example that your product doesn't need.
Connect Prisma
The plugin or connector lets your AI tool work with your Prisma account. Connect the workspace you just set up.
Choose ChatGPT or Claude above to see these steps.
From the session you'll use to build, ask:
Use the Prisma connection to tell me which workspace I'm connected
to and list its databases. Don't create or change anything yet.An empty list is fine for a new workspace. If the connection fails, ask the AI to explain the error and identify the browser step needed to reconnect. Complete that step and repeat the connection check before asking it to create resources.
Know what each piece does
Your AI tool helps turn instructions into changes to the application. The application is the software the user interacts with. Its code describes the screens and the behaviour behind them.
A database keeps information the application needs to retrieve later. Prisma Postgres is a managed PostgreSQL database. Prisma ORM is a way for application code to work with that data; the Prisma ORM quickstart shows the connection between the two.
Hosting runs the application somewhere your friend can reach it. We'll use Prisma Compute for this, including the early preview. Prisma Composer describes the application's services and database connections so Prisma can deploy them together.
These roles help you understand what you're looking at when something goes wrong. A page can load while its database connection is broken. A working preview can also be available only inside your development environment.
Get a starter you can run
Give the AI your brief and the prompt below. It should follow Prisma's application setup guide, run the build in its cloud session, and deploy the starter to your Prisma workspace.
Choose ChatGPT or Claude above to see these steps.
Read https://www.prisma.io/docs/full-stack-tutorial and use the
current Prisma setup instructions to create a web application
for my brief. Choose a supported template with a user interface.
Use Prisma Composer, Prisma Postgres, and Prisma Compute.
Start with the request page and clearly labelled sample content.
Keep the database setup from the starter; we'll connect the
product's complete flow to saved data later.
Run the setup, build, and checks in your cloud environment.
Keep my steps in the browser, with no local software installs
or terminal commands. Explain any browser authorization steps
needed to deploy to my Prisma workspace. Keep credentials out
of chat and source files, and ask before adding a paid service.
Deploy the starter to Prisma Compute. Give me the application URL
and its project in Prisma Console, and verify that the page loads.
Tell me where the code is saved and how we'll update this same app.The Prisma connection lets the AI manage resources in your account. The build tools may also need authorization to upload the application. If that is missing, ask the AI to name the service asking for access, explain what access it needs, and give you the supported browser steps. If your session cannot complete that setup, stop here and record the blocker. A successful connection check alone doesn't show that deployment is available.
Open the Prisma-hosted URL yourself. Ask for one small change, such as changing the heading to “Review the latest draft,” and have the AI redeploy it. Check that the change appears. This early deployment gives you a preview to work with; Get your live version ready to share covers publishing and checking the finished flow before sharing it with a friend.
If you're working in a project with code files, ask the AI to orient you:
Inspect the project before making changes. Explain where the main
screen lives, how to start the preview, and how to run the checks
already included in the project. Identify any data connection that
is already configured without showing secret values.
Tell me which parts of our brief are already supported and which
still need implementation.You don't need to understand every file. You should know where your work is saved and how to tell whether the latest change is visible in the preview. If an instruction assumes software or an account you don't have, tell the AI what's missing and ask whether it can complete the step with your current tools. Check any additional account requirement or cost before continuing.
Keep a way back
Use the project's version history to save the working starter. In a code repository, this is usually a Git commit: a recorded version you can return to. Some building tools expose their own history instead. Ask the AI to identify the method your project uses and show you where the saved version can be found.
Give the checkpoint a useful description, such as “starter runs before feedback flow.” Then ask for a small change to a heading and check it in the preview. Ask the AI to restore that heading from the saved version, keeping the database untouched, and check that the original wording returns. If you're checking the hosted URL, have it redeploy the restored version too. Practise this while the starter contains only sample content, before building the first flow.