AI coding costs are projected to exceed the average developer’s salary by 2028, driven by the increasing consumption of large language model (LLM) tokens and the shift towards consumption-based licensing models, according to a report by Gartner Inc. Organizations transitioning from experimental to widespread use of AI coding agents face significant cost hikes due to token consumption. These AI tokens represent the data processed by generative AI models and directly impact the expenses associated with AI coding tools, especially under consumption-based pricing structures.
Nitish Tyagi, Sr. Principal Analyst at Gartner, highlighted that organizations are swiftly moving towards scaled deployment of AI coding agents but may be underestimating the financial implications of escalating token consumption. He emphasized the need for disciplined token usage, cautioning that without a regulated engineering operating model, costs could outpace the productivity gains intended by these tools. The report noted a shift in AI coding agent vendors from seat-based licensing to consumption-based pricing, leading to variable cost structures for software engineering tasks.
The transition to consumption-based pricing in AI coding tools introduces significant variability in cost structures, with many vendors lacking transparency in calculating and billing token consumption. This lack of clarity hampers enterprises’ ability to forecast and manage costs accurately, potentially resulting in budget overruns and challenges in tracking cost-to-value outcomes. Tyagi pointed out that most organizations currently lack the necessary maturity and frameworks to effectively assess costs against business impact, leading to concerns among software engineering leaders regarding the justification of token-driven AI expenditure.
The report also raised concerns about unregulated autonomy in agent-driven workflows, expansive context windows, and the absence of structured feedback mechanisms that could lead to overspending. It recommended that leaders establish a decision framework based on use cases, align model selection with task complexity, enforce context engineering practices, and implement governance and cost controls to address these challenges.
