Why is there no single AI app development cost?
AI app development cost depends on the same things as any other build, with one change: the typing gets faster while the deciding stays the same. The middle of the build, meaning the screens, form handling, plumbing between services, tests and repetitive parts, is cheaper. The edges, meaning the plan, your data, payments, the store rules and the testing that needs someone who knows your business, are roughly unchanged.
That is not a small difference. It is how a repair estimated at 100+ hours became a 7-day rebuild, live on iOS and Android. AI helped with the typing. It did not write the plan, and it did not know what happens when a booking arrives with no signal.
What does AI change, and what does it not?
This is the split I use when quoting an AI-assisted build.
| The work | What AI changes | What still sets the price |
|---|---|---|
| The plan | Very little | Someone deciding what the app must do, and what it must never do |
| Screen building | A lot faster | How many screens, and how different they are |
| Data model | Some help | Getting the relationships right before anything is built |
| Payments | Boilerplate is quicker | Refunds, failures, payouts and receipts |
| Admin screens | Faster | Who needs to see what, and who may change it |
| Integrations | Some help | What each system expects, and what happens when it is down |
| Testing | Faster to write | Testing the awkward cases, on a real phone and a slow connection |
| The stores | Very little | Apple and Google rules, screenshots, privacy answers and the wait for approval |
| After launch | Very little | Someone to fix it when a feed stops or a payment fails |
How do you price an AI-assisted build?
The same way I price any other build: a free call, a written plan of what I will build and what it will not do, then one fixed price. The price does not move unless the work does.
There is no AI discount and no AI surcharge. If AI saves time, that time goes into testing and into the parts of your business a model cannot see. Small jobs still start at $50.
What does it cost to run an AI feature?
Building it and running it are two different bills. The build is my fixed price. The running bill is the model provider's, and it follows usage.
For most small business features this is dollars a month, not thousands. It becomes a real cost when every user action calls a large model, or when every call sends a long document. I work the monthly number out from published prices and your expected volume before I build, so it is not a surprise.
- How many calls a day, and what each call sends.
- Which model, because the price per token varies widely between them.
- Whether results can be cached instead of asked for again.
- What happens on a retry, so a failed call does not quietly double the bill.
How much does it cost to build an app with Claude Code?
Claude Code is one of the tools I use, not a product you buy from me. Its own subscription or API bill belongs to whoever runs it, and it is small compared with the build.
What you pay for is the plan, the data model, the parts a model cannot decide, the testing and the launch. A price per screen because an agent typed it is either a guess or a quote with the testing left out.
When is AI the wrong way to build it?
Often, for part of the job. These are the cases where I lean on it least.
- Rules that live in someone's head, where the model has nothing to read.
- Anything where a wrong answer moves money and nobody is checking.
- Heavy offline use, where the app has to work with no signal.
- A design that has to be exactly right, screen by screen.
- Old systems with no documentation and no test environment.
How do I compare two AI-assisted quotes?
Send the same brief to both, and compare what is included before you compare the numbers.
- What each quote includes, especially testing and the store submission.
- Who owns the accounts, the code and the data.
- What the monthly running bill is expected to be, and who pays it.
- What the fix-it promise covers after launch.
What to remember
- AI cuts the typing, not the deciding.
- Scope, money, admin, integrations and the stores still set the price.
- The model bill is separate, and small for most business features.
- Ask what is included before you compare two numbers.
Sources
Primary documentation, for anyone who wants to check the details.