All notes Product building

How to build an app with AI: a guide for non-technical founders (2026)

In short

To build an app with AI without coding, name one problem, write a one-page spec, pick an AI app builder (Lovable, Bolt, Base44 or Replit, about $16 to $25 a month on entry plans), build one screen and one flow at a time, and test with five real users. AI gets you a working first version in days. It does not handle accounts, payments, security review or going live on its own. Bring in a developer before the app holds real customer data or takes real money.

A lilac background with a pixel mosaic, a dark prompt card that reads "Build a booking app for my cleaning company", an arrow, and a white app window with a calendar and a green "Book" button

If you want to know how to build an app with AI and you have never written code, the short version is this: describe one problem in plain language to an AI app builder, let it generate a first version, then improve it one small request at a time. A weekend is enough for a prototype you can click through. Getting from that prototype to something customers can safely pay for is a different job, and most guides skip it.

I read the pages that rank highest for this question in the US: Tech.co, Website Builder Expert, Google Cloud, Bubble and Base44. Three come from companies that sell an AI app builder, and two earn a commission when you sign up for one. None of them shows what was left for a human to do after the AI finished.

We've been on both sides: founders who had to validate our own product ideas, and the team that designs and builds products for clients. Max Miskow built his own product, Podium, with an AI coding assistant, and kept a build diary. This guide uses that diary to show where AI saves weeks and where it stops.

What it means to build an app with AI

Building an app with AI means describing what you want in plain language and letting an AI model write the code, the database and the screens. You steer by reading the result and asking for changes. The popular name for this is vibe coding. The person doing it needs a clear idea of the product far more than knowledge of a programming language.

Two families of tools exist.

What is an AI builder? An AI app builder is a website where you type a description and get a running app with hosting included. Lovable, Bolt, Base44 and Replit belong here. You never see a terminal. Lovable describes itself as software that lets you "chat to instantly build websites and web apps, with no technical knowledge needed" (Lovable pricing page).

AI coding assistants such as Claude Code and Cursor work inside a real code project on your computer. They are what professional developers use. In the 2026 Stack Overflow Developer Survey (30,903 responses), 73% of respondents said they use coding assistants and agents daily, and Claude Code (66%) and GitHub Copilot (59%) led among coding agents.

AI app builderAI coding assistant
ExamplesLovable, Bolt, Base44, ReplitClaude Code, Cursor
Where you workIn the browser, in a chatIn a code project on your computer
Hosting and databaseIncluded or one clickYou set them up
Skills you needClear writing, patienceReading code, basic developer tools
Best forPrototype, internal tool, simple MVPProducts that must grow and be maintained
Main riskYou can't judge what was builtYou need a developer to drive it well

How to build an app with AI: 7 steps

To build an app with AI, work in this order: problem, spec, tool, data, small slices, real users, review. The first two steps involve no AI tool at all, and they decide most of the result. A vague idea gives you a vague app, only faster.

  1. Name one problem and one user. "Customers of my cleaning company can't book a recurring visit without calling" is a problem. "An app for my business" is a wish. Write down what the person does today and what it costs you.
  2. Write a one-page spec. List the three to five things a user must be able to do, the data you store (customers, bookings, invoices) and what is out of scope. This page becomes your first prompt. A product roadmap can wait, a one-page scope cannot.
  3. Pick one tool and stay with it. Use the tables below. Switching builders mid-project usually means starting over.
  4. Start with the data, then one flow. Ask for the data model first ("a customer has many bookings, a booking has a date, an address and a status"). Then ask for a single flow from start to finish, such as "a customer books a visit".
  5. Build in small slices. One request, one change, one check. Long prompts that ask for ten features at once produce apps where three of them quietly don't work. Save a version before every larger change.
  6. Test with five real users. Sit next to them and say nothing. Every place they stop is your next request to the AI.
  7. Get a review before real data. Before the app stores customer details or takes payments, have a developer review accounts, permissions, payments and backups. The security section below explains why.
Diagram of seven steps to build an app with AI: problem, one-page spec, tool, data model, small slices, five real users, developer review, with the first two marked as done without AI and the last two marked as work for humans
Seven steps from a problem to a reviewed first version

Bubble, which sells its own builder, puts the limit honestly in its guide: AI can get you about 80% of the way to a working app, and the last 20% often turns out to be the hardest part. Steps 6 and 7 are that last 20%.

Best way to build an app with AI: tools and prices compared

The best way to build an app with AI as a non-technical founder is an AI app builder with hosting and a database included: Lovable, Bolt, Base44 or Replit. Entry paid plans cost $16 to $25 a month. Claude Code and Cursor are stronger tools, but they assume someone on the team can read code.

Prices below come from each vendor's own pricing page, checked on November 20, 2026, with monthly billing in US dollars. All of them change often, so open the page before you pay.

ToolTypeFree planEntry paid planWhat you pay forCode is yours
LovableApp builder5 build credits a day, up to 30 a monthPro, $25/mo, 100 creditsCredits per messageYes, per its pricing FAQ
BoltApp builder1M tokens a month, 300K a dayPro, $25/mo, from 10M tokensTokensYes, GitHub integration
Base44App builder25 message credits a monthStarter, $16/mo, 100 message creditsMessage and integration creditsYes, GitHub export
ReplitApp builder and workspaceStarter: daily AI credits up to a monthly capCore, $20/mo with $20 of AI creditsEffort-based pricing per checkpointYes, GitHub sync
Claude CodeCoding assistantNoClaude Pro, $20/mo ($17 with annual billing)Subscription with usage limitsYes, files live on your computer
CursorCoding assistantLimited requestsPlus, $8/mo, Premium $30/moShared AI usage pool, third-party models billed separatelyYes, files live on your computer

Lovable's Pro price is listed at $0.25 per credit in its documentation, which gives $25 for 100 credits. The pricing page shows local currency outside the US.

Three things in this table matter more than the monthly price.

Credits are the real cost. Lovable publishes examples: changing a button color costs 0.50 credits, adding sign-up and login costs 1.20, a landing page with images 1.70. At around one credit per request, 100 credits is on the order of a hundred requests. Debugging burns credits too, because every "that still doesn't work" is a message.

Pricing models differ. Bolt counts tokens and reads your project files on each request, so large projects cost more per change. Replit charges by effort: a complex build costs more than a simple request, and Replit asks for confirmation before paid work starts (Replit billing docs).

Ownership is fine everywhere, in theory. Every tool above lets you take the code out. Whether a developer wants to continue from that code is a separate question. Ask one early.

Bar chart comparing entry paid plans of AI app tools on November 20, 2026: Base44 Starter $16 a month, Replit Core $20, Claude Pro with Claude Code $20, Lovable Pro $25, Bolt Pro $25
Entry paid plans per month, vendor pricing pages, November 20, 2026

Which tool for which job

You want toStart withWhy
Show an idea to customers or investors this weekLovable or BoltFastest path from a prompt to a shareable link
Build an internal tool for your team (orders, schedules, a dashboard)Base44 or ReplitDatabase, logins and hosting in one place
Sell a product with accounts and subscriptionsApp builder for the prototype, then a developer with Claude Code or CursorPayments, permissions and data need review
Add a feature to software you already haveA developer with Claude Code or CursorExisting code needs someone who can read it
Publish in the App Store or Google PlayCheck store support first, plan for a developerStore accounts, review and updates are manual work

Can you build an app with AI for free?

Yes, you can build an app with AI for free up to a prototype, and no further. Free plans give you enough messages for a first version and a public link. They run out during corrections, and they keep the builder's branding or limits on your published app.

Free planWhat you get (November 20, 2026)Where it ends
Lovable Free5 build credits a day, capped at 30 a monthAbout 30 small changes a month, no custom domain
Bolt Free300K tokens a day, 1M a monthBolt branding on sites, daily limit
Base44 Free25 message credits, 100 integration credits a month25 messages is one short session
Replit StarterDaily AI credits up to a monthly cap, Lite build only"Made with Replit" badge, full builds need Core

A sensible plan: learn how to build an app with AI for free during week one, find out which tool you get along with, then pay for one month of the entry plan. A full first version rarely fits in free limits, because most of the work is fixing, not generating.

How to build an app with Claude or Claude Code

You can build an app with Claude in two ways. In the Claude chat app, you ask for a small tool and get an interactive artifact you can share. With Claude Code, you work in a real project folder, and Claude reads files, writes code, runs it and fixes errors. Claude Code is included in the Pro plan at $20 a month (Claude pricing).

For a non-technical founder the honest advice is this. Use the chat app for calculators, simple forms and one-screen tools. If you want to learn how to build an app with Claude Code, expect a learning curve: you need a code editor, hosting, a database account and a way to keep versions (GitHub). Claude will explain each of these and do most of the typing.

A workflow that works when you build an app with Claude Code: paste your one-page spec and ask for a plan before any code, build one stage at a time with a short test for each, and keep a diary file that Claude updates after every session. Then ask for a list titled "what you must do yourself" (accounts, keys, domain, payments). That list is where the Podium example below comes from.

What AI gets wrong: security and the last 20%

AI writes code that runs. It does not reliably write code that is safe. Veracode tested more than 100 AI models and reported that 45% of the code samples failed security tests and introduced a vulnerability from the OWASP Top 10, the standard list of the most serious web application risks (Veracode 2025 GenAI Code Security Report).

The 2026 edition did not show progress. According to The Next Web's summary of Veracode's 2026 report, the average security pass rate sits at 56%, against 55% at the start, so models still fail about 44% of the time. One caveat from the same article matters: the tests ran on raw models with no human review, so this is the rate at which flaws are generated, not the rate at which they reach production. Review is what closes the gap.

Chart of Veracode results: 45% of AI-generated code samples failed security tests in the 2025 report across more than 100 models, and the average security pass rate was 56% in the 2026 report, meaning about 44% still fail
AI-generated code and security tests, Veracode 2025 and 2026

Professional developers treat AI the same way. In the 2026 Stack Overflow survey, 48% said they trust AI when they can easily validate the answers, and only 16% extend that trust to most tasks. A non-technical founder uses the tool precisely because they cannot validate the code.

What tends to go wrong in AI-built apps, in the order it usually hurts:

  • Permissions. Any logged-in user can see another customer's data by changing a number in the address bar.
  • Secrets. A payment or email key ends up in code that the browser can read.
  • Payments. Checkout works in test mode. Taxes, invoices, refunds and failed cards were never built.
  • Data loss. No backup exists, and a careless request to the AI rewrites the database.
  • The loop. The AI fixes one bug and brings back another, and credits burn while nothing improves.

None of these show up in a demo.

AI app builder or developer: what to choose

Choose an AI app builder alone when the app is a prototype or an internal tool with no sensitive data. Choose a developer who works with AI when the app takes money, stores customer data or has to run for years. For most small companies the best route is both, in sequence: you prototype, a professional rebuilds or hardens what proved useful.

You with an AI builderDeveloper using AIBoth in sequence
Time to first clickable versionDays2 to 3 weeks for a small projectDays
Monthly tool cost$16 to $25 on entry plansIncluded in the project$16 to $25, then project price
Who checks securityNobodyThe developerThe developer, before launch
FitsPrototype, demo, internal toolProduct with payments and customer dataAn idea you want to test before you invest
Main riskIt works until it mattersYou pay before the idea is testedThe prototype may be rebuilt, not reused

If you are wondering how to build an app with no coding experience, or how to build an app from scratch without betting the company on it, the third column is the answer. Your prototype is the best brief a developer can get. It shows screens, flows and data better than any document, and it makes the first conversation much shorter.

This is how we work with companies outside tech. A current client runs several home-service businesses and wants one app where customers can order any of their services, connected to the field-service software the company already uses. Before anyone builds that, the work is deciding what the first version must do and what can wait. If a simpler tool solves the problem, we say so: our guide to a custom app for a small business lists the cases where off-the-shelf software is enough.

Example: what AI did and what a human had to do on Podium

On Podium, the AI coding assistant had first versions of all seven planned stages ready a week before the first stage was due, and the human work took the following weeks. Podium is Max's own product: it tracks how a website shows up in AI answers and writes articles for it. The plan had stages T1 to T7 scheduled from September 9 to October 21, 2026.

Podium dashboard listing keywords with opportunity, difficulty, monthly search volume and article status for each
Podium dashboard, a product Max built with an AI coding assistant

The build diary entry for the night of September 2 records that all seven stages had been checked locally on mock data: a database schema of 20 tables, sign-in, onboarding, the scan engine, the article engine, publishing to three CMS platforms and a weekly email. That is the part every AI demo shows.

The next entries are less glamorous. The first live test on September 3 hit three failures in one evening: one AI provider returned a rate-limit error, another had retired the model the code called, a third had no credits on the account. The first real article cost $2.10 across three attempts, so article writing moved to a cheaper model and each site got a default AI budget of $2 a day. The publishing adapters were, in the diary's words, written from documentation and not yet tested on a live account.

What the AI assistant didWhat a human had to do
Wrote the database schema (20 tables) and migrationsDecide what the first version includes and what waits
Built sign-in, onboarding, scan, article and publishing codeCreate accounts and paste API keys for every provider
Wrote smoke tests on mock data for each stageRun the first test on a real site and read the failures
Wrote payment code for a 3-day trial with a cardSet up the payment account, tax registration and invoice details
Wrote alerting code for failed jobsCreate the alert bot and the monitoring checks
Wrote the deployment scriptPoint the domain, set up production sign-in, set up off-site backups
Proposed model choicesSet the budget after seeing the $2.10 article

The diary has a section titled "What Max must do himself". It lists nine items, from API keys to Google's verification of search data access, which it calls the longest track. None of them is writing code. Alerts were confirmed working on September 16 and live payments were configured and checked on September 17, two weeks after the code "worked".

Timeline of the Podium build: on September 2, 2026 first versions of seven stages pass checks on mock data, a week before the first deadline, on September 3 the first live test hits three provider errors and a $2.10 article, on September 16 alerts work in production, on September 17 live payments are configured, with nine items a human had to do
Code in days, going live in weeks: the Podium build diary, September 2026

The lesson for a founder: budget your time for the right column. AI compressed the typing. Decisions, accounts, money and checking stayed human work.

What an app built with AI costs in the first three months

Three months of building with an AI app builder costs $48 to $75 in entry subscriptions, plus extra credits when you debug a lot. The numbers below are simple multiplication of list prices checked on November 20, 2026. They leave out your own time, which is the largest cost by far.

ScenarioTools3-month tool costWhat it leaves out
Prototype to show customersBase44 Starter$48Extra credits, custom features
Prototype to show customersLovable Pro or Bolt Pro$75Credit or token top-ups
Internal tool for a teamReplit Core$60Usage above the $20 monthly credits
Founder learning Claude CodeClaude Pro$60Hosting, database, domain
Product with customersBuilder for 1 month, then a developer$16 to $25, then a fixed project priceRunning costs after launch

For the last row we give a range after a first call and a fixed price before each stage, so the cost of the build is known before work starts. The MVP section of our home page describes how a first version is scoped.

Running costs continue after launch: hosting, the database, email sending, and AI usage if the app itself calls a model. In Podium, a scan of 22 search phrases cost $0.10.

Checklist before real customers use your AI-built app

Before real customers touch an app you built with AI, confirm ten things. If you cannot answer one of them, that item is the brief for a developer review.

  1. Can user A see user B's data? Test it with two accounts.
  2. Where are the secret keys, and can the browser read them?
  3. Is there a daily backup stored outside the app, and has anyone restored from it once?
  4. What happens when a payment fails or a customer asks for a refund?
  5. Who receives an alert when something breaks at night?
  6. Is the code in your own GitHub account?
  7. Do you own the domain, the hosting account and the database account?
  8. Does the app have a privacy policy that matches what it stores?
  9. Is there a spending limit on every AI and cloud account?
  10. Can you name the one number that tells you the app is working for customers?

To talk through your idea, you can ask for a free report with ideas for your company or book a 30-minute call.

Frequently asked questions

Can I build an app with AI with no coding experience?

Yes. AI app builders such as Lovable, Bolt, Base44 and Replit turn a plain-language description into a working web app with a database and hosting. You need a clear problem, a one-page spec and patience for many small corrections. You will need a developer's review before the app stores customer data or takes payments.

Can I build an app with AI for free?

You can build a prototype for free. On November 20, 2026, Lovable's free plan gave 5 build credits a day (up to 30 a month), Bolt gave 1 million tokens a month and Base44 gave 25 message credits a month. Free limits usually run out during corrections, so plan for one paid month at $16 to $25.

How long does it take to build an app with AI?

A clickable first version takes a weekend to a week. A version that real customers can use safely takes several more weeks, because accounts, payments, backups, alerts and testing are manual work. On Podium, first versions of all seven planned stages passed checks on mock data by September 2, 2026, and live payments were configured on September 17.

Is an app built with AI secure?

Not by default. Veracode's 2025 report found that 45% of AI-generated code samples failed security tests, and its 2026 report showed an average pass rate of 56%. Those tests had no human review. A developer review of permissions, secrets, payments and backups before launch is the practical fix.

How do I build an app and make money from it?

Start from a problem people already pay to solve, not from the app. Charge from the first version, even a small monthly fee, because payment is the only reliable proof of demand. Building the app with AI lowers the cost of testing an idea. It does not create demand.

Who owns the code when I build an app with AI?

You do with the tools in this guide. Lovable states that you own your code and projects, Base44 offers GitHub export of the full source, and Claude Code and Cursor work on files on your own computer. Keep the code in your own GitHub account.

When should I hire a developer instead of using an AI app builder?

Hire one when the app handles payments, stores personal data, connects to systems your business depends on, or has to run for years. Also hire one when the AI starts looping on the same bug. A prototype you built yourself makes that engagement shorter, because it shows exactly what you want.

More notes

All notes