Gartner, an IT market research firm, forecast on June 24 that AI coding costs will exceed the average developer salary by 2028.

It cited rising large language model token consumption and a shift to consumption-based billing as the main reasons.

Nitish Tyagi (니티시 티야기), a senior analyst at Gartner, warned that organisations are rapidly moving AI coding agents from experimentation to large-scale deployment, but often underestimate the financial impact of rising token consumption. He said developers tend to prioritise speed and convenience over cost efficiency, and that without governance token costs can rise faster than productivity gains.

A shift from seat-based, per-user licensing to usage-based billing is also making budget management harder for companies. Many vendors do not disclose token measurement and billing methods transparently, making it difficult for companies to forecast costs accurately.

Tyagi said most organisations lack mature systems to measure business impact relative to cost. He said software engineering leaders are struggling to justify AI spending as budgets run out earlier than expected.

Token overuse is also linked to a lack of agent workflow management, unnecessarily large context windows and the absence of feedback systems to optimise use. Cost-optimisation features offered by AI coding vendors are also still at an immature level.

Gartner urged software engineering leaders to take five steps to manage costs. It said they should set clear standards for AI use by task type, including developer-led, agent-assisted and fully agent-led work. They should select models based on complexity, using small models for simple tasks and frontier models only for complex tasks. They should train developers in context engineering to optimise token use by reducing unnecessary information. They should introduce governance and cost controls. It also said they should review token-heavy work practices in regular development meetings to find inefficiencies and share improvements with the whole team.