Service · Product builds
AI product development
Products where the AI is part of the job
Customer-facing apps on iOS, Android and the web; internal tools that replace a spreadsheet and three inboxes; assistants that answer from your own documents; agents that carry out a defined task and report what they did. Each one ships with the parts that make it trustworthy in use: an evaluation set that tests the AI behaviour, logging you can read, and a way for a person to check or override the output.
We work in your repositories and cloud accounts from the first day, so the code, the prompts and the evaluation sets are yours as they are written.
The code got cheaper. Testing did not.
We quote product builds from the AI-native column of our published cost model. The model counts hours category by category, and the categories behave differently. Code drops sharply, because agents generate it well. Testing across real devices does not move. A new category appears: review and hardening of generated code, which the old way of working folded into build.
Here is the model's default app, both columns, as the engine prints them:
| Category | AI-native | Agency quote | AI-native agency |
|---|---|---|---|
| Design | 84 | 139 | |
| Build (code) | 121 | 399 | |
| Integrations | 74 | 148 | |
| Code review and hardening | 53 | none | |
| Testing across devices | 109 | 109 | |
| Store release | 26 | 26 | |
| Project time | 64 | 82 | |
| Total hours | 531 | 903 |
Across the model's eight example apps, the AI-native total lands at 54% to 61% of the agency total. That range is an outcome of the category tables, not a discount someone chose, which is why we are comfortable quoting from it.
The quote multiplies those hours by a day rate stated on the quote, phase by phase. Running costs sit on their own lines: store fees, hosting, and AI usage at the provider's price. The method page sets out how a quote is assembled, and our five answers cover what happens when a phase finishes early.
The AI bill, estimated before you build
An app with AI features pays a model provider every month. The model estimates that bill as monthly users × requests per user × tokens per request × the provider's published price. For 1,000 monthly users of a chat assistant, it comes to $10 a month on gpt-6-luna, $75 on Gemini 3.8 Flash and $200 on Claude Sonnet 5.5. With search and content generation added as well, the range is $15.80 to $310.30 a month.
We put that estimate on the quote with its assumptions, so the choice of model is a decision you make with the numbers in front of you. The live figures and assumptions are on the model's AI app development cost page.
Questions people ask us
How much does it cost to build an AI app?
It depends on the scope more than on the AI. In our published model, a standard cross-platform app with sign-up, profiles, push notifications, search and analytics comes to 531 hours built AI-native, against 903 hours at typical agency effort. Adding AI features adds their own build hours and a monthly usage bill. Put your own scope into appdevelopmentcostcalculator.com and you will see the hours we would quote from.
What does AI-native development mean in practice?
AI coding agents write most of the code: screens, data handling, wiring, tests. People write the specification, review every change, design the product, test it on real devices and handle release. The review step grows as the code-writing step shrinks, and our quotes show both.
Who pays for the AI model usage once the app is live?
You do, directly. We set up the provider keys in your own account so the bill comes to you at the provider's price. During the build, any usage on our account appears at cost on its own invoice line with the provider invoice behind it.
Can you take over an app another agency started?
Yes, if the code and accounts are yours to share. We start by reviewing the repository and the remaining scope, then quote the remaining work from the same model. If a rewrite would cost less than repair, the review says so with the hours for each.
Start a conversation
Send us a scope, or a link to your numbers in the model. We reply with questions or a quote.