Three different things hide behind “build an app with AI”. Here is what each gives you, one worked example each, and where you still need an engineer.
Ask ten people what “build an app with AI” means and you get three answers dressed up as one: the founder means a product, the ops manager means the tool IT never has time for, the tinkerer means the thing they have always wanted. All three are sold the same demo video.
The adoption is real: in Stack Overflow’s 2025 Developer Survey, 84% of respondents were using or planning to use AI tools and 51% of professional developers use them daily. It also found what the demos skip: more developers distrust the accuracy of AI output (46%) than trust it (33%), and only 3% “highly trust” it. That is not a contradiction; it is the working model: AI builds quickly, a person checks.
So this guide is organised by what you get: three rungs, one worked example each, then the caveat.
| Rung | You give it | You get | Fits | Does not do |
|---|---|---|---|---|
| 1. Custom tool | One prompt | An interactive tool inline in chat; save as a tile, download as HTML, or convert | One job on one screen: calculators, intake forms, mini dashboards | User accounts, shared data |
| 2. Code project | A brief with rules and sample data | A multi-file app in a real code editor: terminal, debugger, live preview, quality loop | Internal tools, prototypes, anything with logic across files | Decide auth, payments or data for you |
| 3. Scaffolded site | A brief and a stack choice | A React, Vue, Astro, Vite, Express or static project with a preview URL and zero-setup static deploy | Launch pages, docs sites, portfolios | Host your database or take payments |
The rungs are not a ladder to climb in order; plenty of useful tools live forever on rung one. Name the rung first — it decides what to ask for and what to check.
Who it is for: anyone with a repeatable calculation, a form people keep getting wrong, or numbers they want on one screen.
The worked example: a quote calculator for a mobile dog groomer whose every quote is five minutes of arithmetic on the phone.
What comes back is not code but the tool itself, running inline in the chat: tap the inputs, watch the total change. If the travel rule is wrong, say so — “travel is free within 15 km” — and it is regenerated with the rest preserved. The Academy course on building a live artifact or custom tool calls this “describe behaviour, not just appearance” — the habit that separates a usable tool from an admirable mock-up.
When it is right, four buttons matter. Save as AI Tile pins it to your dashboard as a small app. Save HTML downloads a single file that runs in any browser. Open in new tab gives it its own window. Convert to hands it on — to a visualisation, a document, or the Code editor once it outgrows one screen.
Limits: one job on one screen — no user accounts, no central store of other people’s data, and rightly so. If the tool needs those, you are on rung two.
Never written a prompt with a finish line — outcome, audience, context, constraints? The free Oppermind foundations course spends a lesson on it and pays back on every rung.
Who it is for: ops people and founders who need an internal tool with logic across several files and want to see it working, not read about it.
The worked example: a roster-conflict checker. An operations lead exports shifts and approved leave every fortnight and finds the overlaps by eye. The brief: upload both, flag anyone rostered while on leave, show the conflicts in a table, export a clean list.
That is a multi-file application — parser, matching logic, results view, export — so it opens as a Code editor project, with a terminal, a debugger and a live preview pane. The agent scaffolds it, installs what it needs in a controlled cloud terminal, builds it and opens the preview; drop in your own files and watch the conflicts appear.
Two things separate this from code pasted out of a chat window. You can see it run: the preview updates as the code changes, the terminal shows every command. And a quality loop judges the result against the target you stated and fills the gaps before handing it over — it checks its own work. When the brief said “handle missing dates”, the loop notices the first draft did not.
You are still the reviewer. Run it against last fortnight’s real files; try a shift with no end time, a leave record with the dates reversed. The Academy’s testing checklist — empty and invalid inputs, unusual values, small screens, error handling, sensitive-data boundaries — is short, and it separates a demo from a tool people rely on.
Limits: rung two builds what you can specify. It will happily generate a login screen; it cannot decide how accounts, permissions and passwords should work in your business, and that decision is the whole job.
Who it is for: founders and small teams who need something on the public internet — a launch page, a docs site, a portfolio — without provisioning anything.
The worked example: a launch site for a bookkeeping service: home, pricing and FAQ pages, looking like a real business by Friday.
Ask for it and choose a stack, or let the agent choose. Oppermind scaffolds real projects — React with Vite, Vue, Astro, Express for an API, plain static HTML and others — with a build that actually runs and a live preview URL to send a co-founder from your phone.
Then the deploy decision, where honesty matters most. Static deploy is the zero-setup option: hosted, permanent, a public URL, and the default. If you already have a Vercel, Netlify, AWS, Azure or Cloud Run account, connect it under Settings and deploy there instead — on your own account and bill; Oppermind is not offering free hosting on somebody else’s platform. Previews are temporary; deploys are permanent.
Limits: a static site is exactly that. Once it needs a database, logins or payments, you are combining rung three with rung two — and the caveat below. If you are building a product with AI inside it, the Developer API quickstart shows how to call Lato 1 from your own backend with a scoped key, which beats embedding a chat window.
“AI app builder” searches lead to the no-code platforms, so here is what two of the best-known publish, converted at A$1.4003 to US$1 (1 September 2026).
Bubble’s web-only Starter plan is US$29 a month on annual billing or US$32 monthly — about A$40.61 to A$44.81 — and Growth US$119 or US$134, about A$166.64 to A$187.64, each with a monthly workload allowance and per-thousand-unit overages. Lovable’s Pro plan starts at US$25 a month for 100 credits — about A$35.01 — and its Free plan grants five build credits a day, capped at 30 a month.
Fair prices for what they are, and what they are is different: Bubble hosts your database, user accounts and workflows, and charges for the processing they consume. Compare that, not the headline. If your app is the business, a platform hosting the whole stack may be right — work that out with the build-versus-buy decision in AI solutions for business.
Oppermind is Free with no card, Starter A$9.95 a month, Pro A$29.99, Pro Plus A$59.99 — one subscription that also carries the documents, spreadsheets, decks, PDFs, images and video around the app. The full comparison lives in the 2026 AI subscription price guide.
All three rungs live in one workspace and hand off to each other: a tool that outgrows one screen converts to a Code project; a project that needs to be public gets a preview URL and a deploy. Everything around the app — the pricing spreadsheet, the launch deck, the product images, the demo video — is made in the same place, in editors that round-trip real .docx, .xlsx and .pptx files. The deck, the model, the site, the video — one prompt away.
Still deciding whether to build anything? Start upstream: make your business AI-ready covers the data, tools and guardrails to sort out first, and how to implement AI in your business is the path from first conversation to everyday use.
Whatever you’re here to make, make more of it.
Production apps with authentication, payments and databases still need engineering judgement. The AI accelerates; it does not replace the person deciding what happens when a payment fails halfway, who may see whose data, or what becomes of the records when a customer leaves. Those are decisions, not code, and a fast builder will implement whichever one you forgot to make.
The speed evidence is more nuanced than the demos, too. In METR’s 2025 randomised study, 16 experienced open-source developers took 19% longer to finish real issues on their own repositories with AI tools — after predicting a 24% speed-up, and still believing afterwards they had been about 20% faster. The lesson is not “don’t use AI”; on unfamiliar code and new tools — most of this article — the gains are real. It is that perceived and measured speed diverge, so measure: write the brief down, test against it, and treat a working preview as a testable artifact, not production approval.
Where the tool is your product, read the build-versus-buy section of AI solutions for business first. If the answer is “build”, build rung one in an afternoon and let it teach you what the real one needs.
Yes, for the first two rungs in this article. A custom tool such as a calculator, an intake form or a mini dashboard comes from a single plain-English prompt and runs inline in the chat, and you refine it by describing behaviour. A multi-file app in the Code editor is built for you and shown running in a live preview; you review it by using it and testing edge cases, not by reading code. Where you will still need engineering judgement is authentication, payments and databases in a production app.
A custom tool is one job on one screen with no accounts and no central data store: it runs in your browser, saves to your dashboard as a tile, and downloads as a single HTML file. A real app has logic across several files, may store data, and needs testing against real inputs, which is what the Code editor with its terminal, debugger and live preview is for. Oppermind can convert a custom tool into a Code project when it outgrows one screen.
Yes. A scaffolded site gets a live preview URL you can share, and a static deploy with zero setup gives it a permanent public URL. If you prefer Vercel, Netlify, AWS, Azure or Cloud Run, connect your own account under Settings and deploy there instead; those deploys run on your account and your bill, not on free hosting supplied by Oppermind.
Oppermind is Free with no card required, then Starter at A$9.95 a month, Pro at A$29.99 and Pro Plus at A$59.99, one subscription that carries the code editor, custom tools and the rest of the workspace. Dedicated app builders price separately: Bubble’s web-only Starter plan is US$29 a month on annual billing, about A$40.61, and Lovable’s Pro plan starts at US$25 a month for 100 credits, about A$35.01, both as published on 2 September 2026.
Third-party prices in this article are the vendors’ published list prices as at 2 September 2026 and change without notice. US-dollar prices are converted at an indicative rate and are not a quote; your final charge depends on the rate at the time of billing, any fees your card issuer applies, and your own GST position. This article is general information about software and is not legal, tax or financial advice. Oppermind is not affiliated with, endorsed by, or sponsored by any product named here; all product names and trade marks are the property of their respective owners. Check the vendor’s current pricing page before making a purchasing decision.
A custom tool from one prompt, a real code editor with live preview, and a site you can deploy — in the same workspace as everything else you make.