AI coding is aggressively becoming part of many software teams, Gartner’s latest warning suggests the economics are changing fast. Gartner predicts AI coding costs will overtake the average developer’s salary by 2028 because of rising large language model token consumption and the shift to consumption-based licensing.
The firm also says many enterprises are scaling AI coding agents without fully accounting for the cost side of the equation.
The cost of AI-assisted coding is no longer simply about software subscriptions. It is increasingly about how many tokens an organization burns while agents read code, generate code, review changes, and iterate through tasks. Gartner defines AI tokens as the data units processed by generative AI models, and says token consumption directly affects pricing under consumption-based models.
“Organizations are rapidly moving from experimentation to scaled deployment of AI coding agents, but many are underestimating the financial impact of rising token consumption,” said Nitish Tyagi, Sr. Principal Analyst at Gartner. “Token discipline will not emerge through developer choice alone, as developers tend to optimize for speed and convenience over cost efficiency. Without a governed engineering operating model, costs can escalate faster than the productivity gains these tools are designed to deliver.”
The problem becomes visible only when AI coding is used at scale. According to Gartner, many organizations are moving from experimentation to broad deployment, but they are underestimating the financial impact of token use.
[ALSO READ: 92% of executives see agentic AI reshaping business operations, but readiness gap remains the real constraint: Report ]
Why AI coding costs are rising
Gartner says the market is shifting away from seat-based licensing toward usage-based pricing which makes software engineering costs less predictable. And when cost depends on token use, price rises with activity, not just per user.
AI coding agents can scan repositories, generate code, handle debugging loops, and produce multiple output passes. Each step consumes more tokens, and developers generally optimize for speed.
Gartner says many organizations lack the methods to track where token spend is going, how it is billed, and whether the business value justifies the cost. Without that visibility, budget excesses become more likely.
According to Gartner, AI coding vendors still have not built mature cost-optimization features into their products, which leaves customers to manage usage manually.
Rising AI coding trend: from helper to operating layer
AI tools are now being used not only for writing code, but also for managing outputs, reviewing changes, and reducing time spent on routine tasks. CIO Dive, summarizing Gartner’s findings, said AI adoption is pushing employees to spend less time writing code and more time managing AI outputs.
In traditional development, the main expense goes to labor. But in AI-assisted development, labor and token spend run in parallel. As use becomes more frequent, the token bill can climb quickly even if headcount stays flat. Gartner says light users are likely to become mainstream users as familiarity and reliance increase, pushing overall spend higher.
[ALSO READ: Is AI Quietly Killing Traditional Software?]
Gartner’s point is narrower and more practical: software leaders need to treat token usage as a budget item that must be governed like cloud spend.
For software teams, the message is that productivity gains do not automatically equal cost savings. AI can speed up development, but if the usage model is poorly controlled, the cost of those gains may be higher than expected. Gartner says organizations need a governed operating model to avoid that outcome.
Gartner’s recommendations
Gartner lays out a set of controls for software engineering leaders. First, it recommends a use-case-driven decision framework that classifies work as developer-led, developer-with-agent, or fully agent-led. Second, it advises matching model size to task complexity, using smaller models for simpler work and reserving larger models for harder tasks.
Third, Gartner says teams should adopt context engineering practices, meaning developers should provide only the information that is relevant, summarized, and necessary for the task. Fourth, it recommends governance controls such as token thresholds, escalation policies, and automated monitoring. Fifth, it wants regular reviews of high-token workflows during sprint retrospectives.
AI coding is spreading because it saves time and expands developer output. But the AI coding costs are changing, and token consumption is becoming the bill that software teams can no longer ignore.
[ALSO READ: [ALSO READ: AI Is Growing Faster Than the World’s Data Centers Can Handle ] ]
