Guides

How to Build an App With AI

Updated 4 August 2026

Ten years ago “how do I make an app?” had one honest answer: find a lot of money or learn to code. That answer is dead. You can now build an app with AI from a prompt, use a no-code app builder or work with an AI coding agent. Here is the 2026 version, without the sales pitch each tool vendor will give you.

First, the question before the question

Nobody actually wants an app. They want the thing the app does: bookings, matches, savings, an audience. So before choosing how to build, name the riskiest assumption in your idea. Is it "will anyone want this?" Is it "will people use it the way I imagine?" Is it "can this even be built?" The right way to make your app depends entirely on which of those keeps you up at night, because the first version's only job is to answer it.

The four routes

1. No app at all (yet). If the risk is demand, a landing page describing the product, with a waitlist or a pre-order button, answers it for under €500. Ten sharp interviews with your target customers are free. This route feels like cheating. It is the most professional move on this list.

2. AI app builders (Lovable, Base44, Replit). Describe what you want in plain English; get a working web app in hours. No code visible, free tiers to start, tens of euro a month after. Perfect for standard shapes: bookings, marketplaces, dashboards, directories. This is where most non-technical founders should build their first prototype.

3. AI coding agents (Claude Code, Codex). An afternoon of setup and a small learning curve buys you real engineering on demand: a genuine codebase you own, with no platform ceiling. Choose this when your idea has a tricky bit the builders refuse, or when you want to understand your own product deeply.

4. Hiring humans (freelancer or agency). €10,000 to €80,000 in Ireland, months of timeline, and the right choice once you have evidence and need scale, compliance or serious integrations. The classic mistake is making this route first when it should be third.

The decision tree

• Risk is demand → route 1. You do not need software to learn whether people want it.
• Risk is behaviour, idea is a standard web app → route 2. Prototype this week.
• Risk is behaviour, idea has a hard technical corner → route 3, pointed only at that corner.
• Evidence in hand, customers waiting, revenue starting → route 4, now worth every euro.

The steps, whatever route you pick

1. Write the idea as one sentence about the user: who does what and why they would care.
2. Name the riskiest assumption and design the smallest build that tests it.
3. Build for days, not months. Ugly is fine; unused is fatal.
4. Put it in front of ten real potential users, not friends, and write down what they do, not what they say.
5. Decide with the evidence: build more, change course, or gratefully kill it for the price of a nice dinner.

Where people get stuck

The tools are no longer the hard part. The hard part is scoping: knowing what to build, what to fake, and what to ignore until later. Founders who skip that thinking build the whole app badly instead of the right slice well, and the €200 experiment quietly becomes a €5,000 wander. If you want that thinking done with someone who does it every week, that is exactly what a Flight Check is for: 90 minutes, and you leave with a one-page plan for precisely your idea.

The €40,000 first build is finished as a sensible default. Make the app the new way: cheaply, quickly, and pointed straight at the question that decides whether your idea deserves to fly.

Already have a working version? The next decision is different. Read I Built an App With AI. Now What? before polishing or launching it.