How to build an app with AI.
You can have a working app by Sunday. Getting one you'd trust with a paying customer takes a bit more thought. Here are the seven steps I'd follow, which tool suits which job, and the bits the AI quietly skips.
FOUNDER-LED · PLAIN ENGLISH · NO TOOL AFFILIATE LINKS
Describe it, build the core, show it to ten people
Tools like Lovable, Bolt, Replit, v0 and Cursor turn a plain-English description into a working app faster than most people expect. That's the easy bit, and it's genuinely impressive.
The hard bit hasn't moved. It's deciding what to build first, and knowing what the tool left out. Most of the trouble I see comes from skipping that second part, not from the AI getting the first part wrong.
Which AI tool suits which job
These tools change monthly, so treat this as a rough guide. Pick one, build something small, and don't spend a week choosing.
| Tool | Good for | Watch out for |
|---|---|---|
| Lovable | A first web app from a chat, with login and a database set up for you | Who can see which data is still down to you to check |
| Bolt | Chat-to-app in the browser, quick to get something on screen | A big pile of code that nobody has read |
| Replit | Building, running and hosting in one place | Your app and your hosting share one account, so know whose it is |
| v0 | Screens and interface pieces | It's mostly the front end. You still need the back end |
| Cursor | An editor with AI built in, for someone who can read code | It'll cheerfully do exactly what you asked, including the wrong thing |
Seven steps, in this order
Write the one-sentence brief
Who is it for, and what does it do for them? If it takes a paragraph, the idea is still too big. Everything you prompt later hangs off this sentence.
Sketch the journey on paper
Draw the three or four screens someone passes through to get the result. Ten minutes with a pen saves hours of re-prompting.
Build the core first, one feature per prompt
Get the thing at the centre working before you add anything else. Small prompts are easier to check, and easier to undo when the AI goes sideways.
Put the code in your own GitHub from day one
Most of these tools can sync to GitHub. Switch it on now. If the tool vanishes or you fall out with it, your code doesn't go with it.
Set the data rules before you add real data
If you're on Supabase, the rules that decide who can see what need setting up properly, and AI often leaves them wide open. The Supabase security checklist takes about ten minutes.
Try it with ten real people
Not friends who'll be nice. People who have the problem. Watch where they get stuck, and what they ask for next. That's your roadmap.
Know when to stop prompting
When a paying customer, real payments or someone else's data are in sight, change from "does it work?" to "is it ready?". That's a different question, and the next section is about it.
It works in the demo. That's not the same as ready.
AI tools are brilliant at making something look finished. They're much less reliable at the dull, invisible bits: who owns the accounts, what happens when a payment fails, whether there's a backup, and whether one customer can see another's data.
None of that stops you launching. It just tends to turn up at the worst moment, usually when an enterprise buyer's security team starts asking questions.
You can test a lot of it yourself. Ten checks you can do yourself walks through them in plain English. If it's already breaking with real users, see how to fix a vibe-coded app.
The subscription is the cheap bit
The tool subscriptions are small, and you'll pay for hosting and a domain on top. The real cost is your time, and what it takes to put right anything the AI got wrong that you only find out about later.
If you'd rather not spend your weekends prompting, I build the real working core as production code in 2-5 days, from £1,500. The UK MVP cost guide has the full prices.
Building an app with AI, answered
Can I build an app with AI if I can't code?
Yes, up to a point. You can get a working first version without writing code. Where non-coders get stuck is what comes after: who can see what, where it's hosted and who actually owns it. That's where a second pair of eyes earns its money.
Which AI tool is best for building an app?
It depends on the job. Lovable and Bolt suit a first web app, Replit keeps building and hosting in one place, v0 is strong on screens and Cursor suits someone who can read code. Pick one and build something small. Switching tools costs more than the differences between them.
How long does it take to build an app with AI?
A rough prototype can appear in an afternoon. A version you'd let a paying customer use takes longer, because the unglamorous parts take the time: access rules, error handling, hosting and testing. How long an MVP takes has realistic timelines.
Is an app built with AI safe to launch?
It can be, but don't assume it. AI tools build what you ask for and tend to skip what you didn't think to ask. Run the ten readiness checks before real customers or real data go near it.
Do I own an app I built with AI?
You should, but check. The code should sit in a GitHub account you control, and the hosting, database and domain accounts should be in your name. If any of them are in the tool's account or someone else's, fix that first.
Related guides
Is your AI-built app ready for customers?
Ten checks you can do yourself, in plain English.
Read the guideSupabase security checklist
Ten minutes to find out if other users can read your data.
Read the checklistBuilt something already?
Let's talk. Bring what you've got and tell me what's coming next, like a first customer or a pilot. I'll tell you what's fine, what's risky and what can wait.
Not built anything yet? See how I build an MVP from scratch.
Darryll Wolloff is the founder of TechConnex. He builds MVPs and fixes AI-built apps for early-stage founders, and still writes and ships production code himself. Last updated 18 November 2026.