How agencies price app development in the AI era

1. Rate cards have not moved; hours have

The published hourly bands we read on 2 October 2026 sit 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, after single-digit falls in 2025.

Freelance rates show the same flatness. freelancermap's Freelancer-Kompass 2026, a survey of 5,412 freelancers between November 2025 and February 2026, found an average hourly rate of 103 euros against 104 euros the year before, the first time in the survey's history that rates did not rise. 62% said they did not plan to raise their rates. That survey covers freelancers in all fields, not only developers.

Meanwhile the hours a task takes are changing, unevenly and in ways the research still argues about (see how AI changed app development costs). Flat rates and changing hours mean the billing model decides who ends up with the difference.

2. Who keeps the speed-up, by billing model

Fixed price at the prototype end

Productised offers with published prices now cover the smallest builds. BuildMVPFast, for example, lists MVPs from $1,999 delivered in 14 days and from $2,500 in 21 days. At that price the scope is necessarily narrow: our smallest worked example, a one-platform validation app, comes to 295 hours, or $9,400 even at Accelerance's Asia rates. Compare what is in the package line by line before comparing prices.

Hybrid contracts

Riseup Labs, a development company, lists fixed price, time and materials, usage-based, value-based, subscription, licence, outcome-based and hybrid models, and says "Hybrid models are increasingly common and well-suited to large or evolving AI solutions." That is a vendor's view rather than survey data, but it matches what a buyer should want: a fixed price where the scope is clear, and hourly billing with a cap where it is not.

The AI specialist tier

Two published sources put AI and machine-learning work above general development rates: Riseup Labs lists AI development at $100-250+ an hour in North America and $90-200 in Western Europe, and the CoderCops blog lists AI/ML specialists globally at $100-250 an hour, citing Index.dev and Arc.dev. Neither publishes its method. If a quote prices AI work at a premium, ask which tasks need that skill and for how many hours.

3. Outcome pricing is growing, slowly

The billable hour is under discussion in professions that sell time, but it is not gone. Writing about management consulting, Marin Ivezic of futureofconsulting.ai describes the shift to outcome-based pricing as "sluggish" and cites one late-2025 analysis finding that only about 25% of McKinsey's fees globally are linked to outcomes. We could not open that underlying analysis, so treat the figure as his summary.

The law gives a parallel from another hourly-billed profession. Thomson Reuters' legal blog describes a "client value squeeze" and argues that when firms use AI to do in minutes what once took hours, "clients reasonably expect to share in those efficiency gains", while noting that 90% of legal spending still flows through billable-hour arrangements.

For an app buyer, the practical form of outcome pricing is modest: part of the fee tied to a launch date, an acceptance test or a performance target written into the contract. It works only where the result can be measured without argument.

4. AI running costs on your invoice

If your app has AI features, it pays a model provider for every request, every month. In our calculator, a chat assistant used by 1,000 people a month comes to about $75.00 a month on Google's Gemini 3.8 Flash at published prices, and much more on larger models (see the model price table). How that cost reaches you varies. Digiday's March 2026 reporting on marketing agencies found all of these in use:

• Passed through at cost. Merge passes token costs to clients "on a metered, case-by-case basis"; Big Spaceship treats compute as a budget line "akin to a catering or equipment hire."
• Absorbed as overhead. RPA "absorbs its clients' token costs", which means they are somewhere in the fee instead.
• Marked up or bundled. Resold AI capability carries a margin on top of the provider's charge. Infracost's glossary lists the common structures: a fixed multiple of token cost, a fixed price per output, cost-plus with a floor, volume bands, or pass-through plus a separate platform fee.

Then compare the answers with an independent number. Price the same scope in the calculator, with every rate and hour assumption visible, and ask about any line where the quote differs by a wide margin.