AI software development is a key focus for providers of Custom Software It’s long been part of everyday life. When a quote is calculated today, there’s a high probability that a developer is sitting at a screen, using AI tools to generate code. For clients, this is good news at first glance, but it raises the question of who ultimately reviews the code and whether this actually affects the price. This article breaks down what’s actually changing, beyond the announcements made by individual providers.
AI-driven software development does not automatically lower the cost of custom software. Some of the time saved is reinvested in testing the generated code, and fixed prices depend on scope, risk, and liability anyway—not solely on developer hours. For clients, two things in particular change: the acceptance process and the question of who is liable for errors in the generated code. Both of these should be included in the contract.
How Widespread Is AI in Software Development?
The Google's DORA Report 2025 surveyed approximately 5,000 software development professionals on this topic and provides one of the most comprehensive datasets on the subject. According to this report, the use of AI tools in development teams will already be widespread by 2025, regardless of company size or industry. In a 2024 forecast, Gartner predicts that by 2028, approximately 75 percent of software developers in companies will be using AI assistants. According to Gartner’s figures, this share was still below 10 percent in early 2023. This trend also affects external service providers who build custom software for clients in industry and retail.
Recent figures from Gartner for the year 2026 show just how significantly this affects day-to-day work. According to these figures, 90 percent of the engineering managers surveyed report improvements resulting from the use of AI, with a reported net productivity gain of around 19 percent. To clients, this may initially sound like a simple math problem—after all, higher productivity should mean lower prices.
This is the reported net productivity gain from AI tools, averaged across teams with widely varying levels of experience.
Gartner, 2026
However, both figures—the 75 percent from Gartner’s 2024 forecast and the 90 percent from Gartner’s 2026 survey—describe the prevalence of AI tools, not their impact on the price of an individual offering. This distinction is important for clients because the two figures are often conflated in sales discussions.
How AI Software Development Is Changing Things for Clients
In practice, the math isn’t quite that simple—for several reasons. First, developers„ time is shifting from purely writing code to reviewing what the AI suggests. Consequently, the effort required for review does not decrease to the same extent as the time spent writing code. Second, fixed prices are usually tied to the scope, risk, and responsibility that a provider assumes—and hardly to developer hours alone. Third, the use of AI requires extra care during the acceptance process because generated code comes with its own set of errors. These are described in detail in the article on the “Risks of AI-Generated Code.”.
In concrete terms, this means that contracts can clarify the use of AI without prohibiting it. It makes sense to have a provision that specifies in which areas AI-generated code may be used, how it is reviewed, and who is liable in case of doubt. A client who asks about this when comparing bids will receive more reliable answers than one who relies solely on hourly rate calculations. This provision is particularly valuable for multi-year projects with rotating developers, because it applies regardless of who on the team actually ends up working on the code.
| Area | A Common Expectation | What Happens During the Cost Estimation Process |
|---|---|---|
| Price | Decreases in proportion to the productivity gain | Some of the time saved is spent reviewing the generated code |
| Fixed Price | Calculated based on developer hours | Depends on the scope, risk, and level of responsibility assumed |
| Acceptance | Remains unchanged | Requires additional testing steps for generated code sections |
| Contract | Doesn't have to mention AI | Better defines the scope of application, testing, and liability |
The productivity gains are unevenly distributed
The Gartner figures for 2026 represent an average that obscures the reality faced by individual teams. A study published in August 2026 in the trade journal Science A recently published study paints a more nuanced picture. According to this 2026 study, the productivity gains from AI tools are distributed unevenly between experienced and less experienced developers. Experienced developers benefit significantly more because they can assess and correct AI suggestions more quickly. Entry-level positions, on the other hand, are coming under pressure because some of the tasks that new professionals used to learn through hands-on experience are now being handled by AI.
Clients would therefore be wise to take a look at a provider’s team composition. A team consisting exclusively of entry-level professionals that relies heavily on AI support has a different risk profile than a mixed team with experienced leadership. Both models can work; they are simply suited to different types of projects.
For clients, this also has longer-term implications. A service provider that no longer allows new hires to learn the ropes through simple tasks—because AI is taking over those tasks—will find it harder to develop experienced young talent in the medium term. This affects clients even if the current quality of projects isn’t yet affected, because the overall availability of experienced developers on the market may become scarce.
What Clients Should Look for When Comparing Quotes
It’s easy to gauge how seriously a provider takes AI software development during an initial meeting, even without requiring them to disclose internal processes. It is revealing to see how AI-generated code is tested before acceptance—by whom and using which tools. Equally revealing is the composition of the team, particularly the ratio of experienced to less experienced developers, as this allows one to assess the risk profile of the proposal. Anyone who receives only evasive answers to such questions should take this into account when comparing prices, regardless of how affordable the quoted hourly rates may seem. This article explains how the overall price for custom software development is broken down. Costs of Custom Software Development.
The Gartner figures and the DORA report describe the prevalence of AI tools, not their impact on the price of an individual proposal. An average increase in productivity cannot be used to derive a discount for a specific project. There are currently no reliable figures showing the extent to which quotes for custom software have changed. Anyone who expects a fixed savings today is relying on an assumption.
Clarify Questions for Comparing Quotes
How generated code is tested prior to acceptance and what the ratio of experienced to less experienced developers is within the team can be specifically addressed during an initial consultation without requiring a provider to disclose its internal processes.
Frequently Asked Questions
Will this make software cheaper?
Not automatically. According to Gartner’s 2026 figures, the reported net productivity gain is around 19 percent, but when calculating a quote, part of that is offset by the additional testing effort. Where prices are falling, this tends to apply to standardized tasks requiring little specialized knowledge, such as recurring form templates or simple data exports. Custom software with complex business logic benefits more slowly because verifying the generated code accounts for the majority of the time spent on such projects, and this verification ties up experienced developers.
Does the use of AI have to be specified in the contract?
This is not strictly required, but it is recommended. A clause specifying the extent to which AI tools may be used, how the results will be reviewed, and who is liable in the event of errors provides clarity for both parties. Without such a provision, it remains unclear whether a security vulnerability in AI-generated code is considered a standard defect or a gross breach of duty by the provider, which can make a significant difference in terms of liability and warranty in the event of a dispute.
How do you check the quality?
Using the same methods as with any other code: automated testing, code reviews by experienced developers, static analysis tools, and—for security-critical systems—a separate review of dependencies and licenses. The toolkit remains largely the same. What has changed is the scope, because AI generates code faster than a team can review it responsibly without additional verification steps. Clients can have these verification steps outlined during the initial consultation and include them in the contract as part of the acceptance process.
torck uses AI tools in its own development, but the generated code is always reviewed by experienced developers on the teams in Maxhütte-Haidhof, Vienna, and Rabat. Because the contractual partner is the German company torck GmbH, the use of AI in offerings for industry and retail can be clearly defined in the contract. Anyone who wants to know how this affects a specific project should discuss it in a Initial Consultation.
German 
English 
French