How to build an app with AI without coding, and know what it can't do
Describe one small, specific tool: who uses it, what they put in, what they get out. Let the AI build it, then test it like a stranger would, on a phone and a laptop, with silly inputs. Before you rely on it, find out where it keeps data and what it can't reach. Many AI-built apps run entirely in the visitor's browser, which suits calculators, games and trackers but not shared or live data.
Pick an app that fits
AI builders are good at self-contained tools: a quote calculator, a quiz, a habit tracker, a game, a planner, a converter. They struggle with anything that depends on a server you'd normally build: user accounts, data shared between people, payments inside the app, live prices or stock, messaging. Decide which kind yours is before you start, because it changes whether you're building the real thing or a prototype.
| Works well | Works as a demo only | Needs a server |
|---|---|---|
| Calculators and estimators | Dashboards with sample data | Logins and personal accounts |
| Quizzes and personality tests | A booking flow to show investors | Data shared between users |
| Games, puzzles, timers | A marketplace mock-up | Live data from other services |
| Personal trackers and planners | A chat interface | Payments inside the app |
Describe it so the first build is close
- Name the user and the moment: "A dog groomer's customer, on their phone, choosing a service before booking."
- List the inputs: "dog size, coat type, services wanted".
- Say exactly what comes out: "an estimated price range and a time, with a button to send the details".
- Give the rules: real prices, formulas or scoring, in plain numbers. Don't let the AI invent them.
- Say what to remember: "keep the last estimate if they come back".
- Say the look in a few words: calm, playful, technical, and any colour you need to match.
Test it like a stranger
- On a phone first. Tap every control with a thumb; check nothing sits under the keyboard when typing.
- With odd inputs. Zero, negative numbers, very long names, empty fields, pasting text into number boxes.
- After a reload. Does what it should remember come back? Does what shouldn't stay away?
- In a private window. That's what a first-time visitor sees, with nothing saved.
- With the rules checked by hand. Work through three cases with a calculator and compare. AI gets arithmetic rules subtly wrong more often than layout.
Know where the data lives
Browser-only apps usually save to localStorage. MDN describes it as storage per origin that persists across sessions, except in private browsing, where it's cleared when the last private tab closes. In practice: each visitor's data stays on their own device and browser, you can't see it, it doesn't sync to their other devices, and clearing site data deletes it. That's fine for a personal habit tracker. It's not fine for anything you need to keep, share or report on.
Building an app with Anticode
Describe the app to Anticode and its Architect talks it through first, asking up to three questions where the answer changes the build, then shows a short plan. Model-written builds need an account. The model writes the whole app as one HTML document (HTML, CSS and JavaScript), streamed as it goes, and the app passes its own code gate with one automatic repair pass. When Cloudflare Browser Rendering is available, a new app build is opened in headless Chrome at phone and laptop widths, screenshotted and clicked through, and the Critic fixes what it finds. That catches a lot; it doesn't replace your own test of the rules.
The app runs in a strict sandbox with no network of any kind. It can save to the visitor's browser, trigger downloads of files it makes, open the print dialog, go fullscreen, use a game controller, play sound it makes and copy to the clipboard. Contact, enquiry, booking and newsletter forms inside it reach you through Anticode's form pipeline, and content you edit, such as a menu, can come from the project's CMS collections, read-only. Anything that would need a server is simulated with sample data and labelled "Demo data". The app sandbox page has the details.
Limits
Anticode apps have no real user accounts, multiplayer, live third-party data, payments or AI inside the app, and no camera, microphone or location. There's no server-side database inside an app. If the tool proves itself and needs those things, the Next.js export (which covers app builds too) gives a developer a standard codebase to continue from; see when to move to custom code.
Questions people ask
Can an AI-built app have user accounts?
Not in a browser-only sandbox. Accounts need a server and a database. Some builders offer them; in Anticode, real accounts exist only in the Next.js export.
Can the app work offline?
A browser-only app needs no server once loaded. Anticode sites also have an offline fallback as installable apps, but test the specific app offline before promising it.
Do I own the app's code?
With Anticode you can export it as a single HTML file or a Next.js project. Check any builder's export options before you invest time in one.
Sources
- MDN: Window.localStorage, checked