Enterprises may soon be paying as much for their developers’ AI token usage as they do for their salaries.

According to Gartner, these costs will meet, or even exceed, the typical software engineer’s monthly salary within the next two years.

This is not only because developers are increasingly adopting generative AI and agentic tools, it reflects a trend toward consumption-based licensing models as vendors balance infrastructure investments with profitability. Rather than the flat per-seat SaaS model of the past, enterprises now pay for developer token use as well.

Gartner senior principal analyst Nitish Tyagi explained that it’s important to note that Gartner’s prediction is based on a global average salary of $2,000 per month; it doesn’t mean AI token usage will exceed all salaries. For instance, in the US, yearly pay rates can be six digits or more.

However, that kind of spend is not out of the realm of possibility, Tyagi emphasized. “I have heard scary numbers like ‘My developer consumed $20K last month,’ or ‘A business user consumed $32K’.”

If these amounts sound shocking, that’s the point. “The goal is to alarm the industry about the impact of token cost if it is not governed and controlled,” he said.

Lack of visibility, immature oversight

Enterprises are quickly moving from experimentation to scaled deployment of AI coding agents, but many still underestimate token costs, Tyagi noted.

This is because cost structures for software engineering workloads are “highly variable,” he pointed out, and there isn’t a lot of transparency into how token consumption is calculated and billed.

AI coding vendors have yet to deliver “mature, built-in cost optimization capabilities,” Tyagi said, and prices will likely only continue to rise as vendors further build out their models while at the same time trying to remain profitable.

Thus, enterprises struggle to forecast and control costs, and, because AI is moving so fast, many organizations lack the “maturity and frameworks” to determine ROI, he noted. Agent-driven workflows are difficult to govern, context windows become bloated, budgets are wiped out earlier than anticipated, and token spend becomes hard to justify.

Added to this, light users such as non-developers will increase their usage as they become more familiar with, and even reliant on, AI tools, driving up token consumption and spend even more.

Tyagi said that, while AI is incredibly valuable, he sees no “direct relationship” between the number of tokens developers consume and their productivity gains. Rather, applying context engineering principles to optimize or reduce token consumption increases quality.

“Tokenmaxxing is not directly related to higher productivity gains,” Tyagi said, “but optimizing token consumption is.”

Still, this in no way means that organizations should move away from AI coding agents, he emphasized. Optimizing token consumption simply means spending only as much as needed without compromising the quality and value brought by AI.

“Without a governed engineering operating model, costs can escalate faster than the productivity gains these tools are designed to deliver,” Tyagi said.

How enterprises can control token usage

The traditional ‘lines-of-code-written’ productivity metric no longer applies when AI can almost instantaneously produce entire Python libraries. Rather, value should be measured in quality, speed, and customer satisfaction metrics, Tyagi said.

For instance: How quickly are developers able to release important features? How much time is reduced between app development and feedback from business, product, and development teams? Shipping features quickly while maintaining quality can create competitive advantage and improve user and customer experience, he said.

Gartner also advises establishing strong governance and cost controls. For instance, introduce token thresholds, automate usage monitoring, and create explicit escalation policies.

“Embedding these controls into engineering workflows ensures consistency and prevents uncontrolled cost growth,” the firm notes.

In addition, enterprises should create a “use case driven” decision framework. This means clearly defining when AI coding agents should be used, and their appropriate levels of autonomy given certain tasks. Further, classify those tasks into three execution models: ‘developer‑led,’ ‘developer‑with‑agent’, and ‘fully agent‑led.’

Enterprises should also select models based on task complexity. Break work into smaller tasks that can be performed by smaller models, “with escalation only when complexity demands it,” Gartner advises. Engineering teams should route workflows deliberately, directing simpler, high-frequency tasks to smaller models and using frontier models only for complex and high-value work.

Another cost saving tactic is mandating specific context engineering practices, the firm says. Developers should be trained to optimize the context they input to AI, including only the information that’s relevant, summarizing that content as much as possible, and eliminating unnecessary data.

Further, teams should embed token usage reviews into development cycles. Regular review of high token consuming workflows can help identify inefficiencies, refine practices, and support collaboration, Gartner says.

Tyagi noted that developers tend to optimize for speed and convenience rather than cost efficiency, so token discipline cannot be achieved through developer choice alone.

His advice for leaders: Do not treat escalating AI coding costs as a reason to move away from AI, or to shift to open generative AI models for everything. “The goal is always to optimize costs without compromising the value.”

Start small, and focus on context engineering first, he said. Assess your current software engineering maturity and select the appropriate agent autonomy. AI assistive development can provide up to 20% productivity gains, “which is not a bad number.”

For developers, he advises: “Target context engineering as one of the most important skills for yourself. This is not only going to help your employer, but also your career.”

This article originally appeared on CIO.com.