How AI changed app development costs (2026)

1. The productivity evidence contradicts itself

Three of the most cited sources, read in full on 2 October 2026:

• GitHub, 2022. GitHub recruited 95 professional developers, split them randomly, and timed them writing an HTTP server in JavaScript. Developers with Copilot completed the task 55% faster (1 hour 11 minutes against 2 hours 41 minutes), with a 95% confidence interval of 21% to 89% for the speed gain. One well-defined task, written from scratch.
• METR, July 2025. A randomised trial with 16 experienced open-source developers working on 246 real issues in repositories they had contributed to for years. With AI tools allowed, they took 19% longer to complete issues. They had expected a 24% speed-up, and afterwards still believed they had been sped up by 20%.
• METR, February 2026 update. METR now says those early-2025 results no longer reflect current tools. Its follow-up reports that developers increasingly declined to take part because they did not want to work without AI, and concludes it is likely developers are more sped up in early 2026 than in early 2025, while stating that its data "is only very weak evidence for the size of this increase." The original 19% figure had a confidence interval of +2% to +39%.
• DORA, 2024. Google's DORA research found a 25% increase in AI adoption associated with better documentation quality (7.5%), code quality (3.4%) and code review speed (3.1%), but also an estimated 1.5% decrease in delivery throughput and a 7.2% reduction in delivery stability.
• DORA, 2025. From nearly 5,000 respondents: 90% use AI at work, and this time AI adoption showed a positive relationship with delivery throughput, but it "does continue to have a negative relationship with software delivery stability." 30% report little or no trust in AI-generated code. DORA's summary: AI "amplifies what's already there."

Put together: a large speed-up on a scoped greenfield task, a slowdown on real work in mature codebases in early 2025, a likely but unmeasured improvement since, and survey data where faster delivery comes with less stable delivery. None of that supports a single "AI saves X%" figure for an app budget, so this site does not publish one.

2. Rate cards did not collapse

If AI had cut the cost of building software sharply, the clearest place to see it would be what development firms charge. The published bands we read today are still in familiar ranges: Clutch lists US development companies at $50 to $99 an hour and app development companies overall at $25 to $49. Accelerance puts senior developers at $60 to $75 in Latin America and $64 to $76 in Europe.

Rates have drifted down, not fallen away. Accelerance reports declines of 7.1% in Latin America, 4.4% in Europe and nearly 8% in Asia over 2025, and names competition, automation and AI-augmented developers among the causes. That is a single-digit price movement, not a step change. All the rates are on the hourly rates page.

3. Prototypes got cheap; production did not

AI app builders turn a description into a working prototype, and their published prices are low. Bolt lists a free plan and Pro at $25 a month (Teams at $30 per member). Replit lists Core at $20 a month ($18 a month billed annually) and Pro at $100.

That changes the cost of finding out whether an idea is worth building, which matters: a clickable prototype used to need design and development time. It does not remove the work a production app needs and that this calculator prices: a backend that holds real users' data, testing across devices, store submissions, and upkeep as Apple raises its minimum Xcode and SDK each spring and Google Play raises its target API level each August (see maintenance cost). A prototype and a production app are different products with different budgets.

4. What AI added to budgets: the model API bill

The clearest cost change is new spending. An app with AI features pays a model provider for every request, every month. Unlike the productivity question, this can be priced from published numbers. For 1,000 monthly active users of all three AI features in the calculator, using each provider's published price read today and our usage assumptions:

The spread between models is wide: about 20 times between the cheapest and dearest rows for the same usage. Google lists its Gemini 3.8 Flash price as rising to $1.50 input and $7.50 output from 1 January 2027, which would roughly double that row. Usage is the other lever: these bills scale linearly with users and with how much context each request carries.

AI features also cost build time. Our allowances (labelled as our methodology) are chat assistant 112 hours, semantic search 72 hours, content generation 64 hours, before platform multipliers, testing and project management. Adding just the chat assistant to our booking app example takes it from $84,000 to $96,000 at the Clutch US band midpoint.

What this means for your budget

1. Do not discount a quote by a headline AI percentage. Ask the team how they work and compare hours line by line.
2. Use AI builders to test an idea cheaply before paying for production work.
3. If the app itself uses AI, budget the API bill per user per month, and re-check provider prices before launch.

Who keeps the time AI saves depends on the contract. See how agencies price app development for the billing models and five questions to ask before you sign.