According to Gartner's market researchers, increasing token consumption and the shift to consumption-based billing models will make coding with generative AI increasingly expensive. By 2028, costs are expected to exceed the global average developer salary; the forecast is based on a worldwide average of around US$2,000 per month.

"Organizations are rapidly moving from experimentation to scaled deployment of AI coding agents, but many are underestimating the financial impact of rising token consumptionn," explained Gartner analyst Nitish Tyagi. However, more discipline in token consumption will not solely arise from developers' decisions, Tyagi elaborated. In his view, they tend to favor convenience and speed over cost-efficiency. Without a regulated operating model for development, costs in companies could rise faster than the productivity gains that AI tools are intended to achieve.

No ROI, nowhere?

"Software engineering leaders are increasingly concerned as token-driven AI spend becomes harder to justify, with budgets often being depleted earlier than expected," said Tyagi. Among others, Uber's CTO Praveen Neppalli Naga caused a stir in April with the statement that the company's annual token budget had already been depleted. Following this, Uber President Andrew Macdonald added in May on a podcast that the benefit of using AI was also unclear. An increase in useful functions for consumers had not materialized. A similar picture is emerging in Germany, where, according to a Bitkom survey, around a third of the companies surveyed have been surprised by the costs of their AI use.

According to Gartner, providers also lack transparency in calculating and billing token consumption. Providers have not yet provided integrated cost optimization functions in their AI coding agents. This makes it more difficult for companies to accurately forecast and control costs.

Transparency and control are lacking

In addition, there is a lack of control over usage within companies, which leads to excessive spending. Gartner cites uncontrolled autonomy in agent-driven workflows and overloaded context windows as frequent sources of error. Overall, the price spiral is likely to continue, Tyagi estimates: "AI coding costs will continue to rise as infrastructure investment and profitability challenges push model pricing higher."

To keep costs under control, Gartner analysts recommend, among other things, introducing token thresholds and automated monitoring. Tasks for AI should also be segmented as much as possible so that they can be handled by smaller models. High-end models should only be used for complex, high-value tasks. Furthermore, developers should be trained to optimize their AI prompts for economy by including only relevant information and summarizing content where possible.

(axk)