Enterprises may soon find themselves 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 prediction is based on the growing adoption of generative AI and agentic tools, combined with a shift toward consumption-based licensing models as vendors balance infrastructure investments with profitability.
Gartner senior principal analyst Nitish Tyagi explained that the prediction uses a global average salary of $2,000 per month, meaning it does not imply token costs will surpass all salaries. In high-paying regions like the United States, where annual salaries can reach six figures, the relative cost may be lower, but the absolute spending can still be eye-opening. Tyagi noted hearing reports of individual developers consuming $20,000 in tokens in a single month, and business users consuming even more.
The goal of such alarming figures is to shock the industry into recognizing the potential impact of token costs if left ungoverned. As enterprises move from experimentation to scaled deployment of AI coding agents, many still underestimate these costs. Cost structures for software engineering workloads are highly variable, and vendors have yet to provide mature, built-in cost optimization capabilities. Pricing is likely to increase as vendors continue to develop models while striving for profitability.
Lack of visibility and immature oversight
Enterprises struggle to forecast and control costs because AI is evolving rapidly. Many organizations lack the maturity and frameworks to determine return on investment. Agent-driven workflows are difficult to govern, context windows become bloated, budgets are exhausted earlier than anticipated, and token spending becomes hard to justify. Additionally, light users such as non-developers will increase their consumption as they become more familiar with AI tools, further driving up costs.
Tyagi emphasized that there is no direct relationship between the number of tokens developers consume and their productivity gains. Instead, applying context engineering principles to optimize or reduce token consumption increases quality. He coined the term 'tokenmaxxing' to describe excessive token use without corresponding productivity benefits. Optimizing token consumption means spending only as much as needed without compromising the value that AI brings.
How enterprises can control token usage
Traditional productivity metrics like lines of code written are no longer relevant when AI can instantly generate entire libraries. Value should now be measured in quality, speed, and customer satisfaction. For example, enterprises should track how quickly developers release important features and how much time is reduced between development and feedback from business teams. Shipping features quickly while maintaining quality creates competitive advantage.
Gartner advises establishing strong governance and cost controls. This includes introducing token thresholds, automating usage monitoring, and creating explicit escalation policies. Embedding these controls into engineering workflows ensures consistency and prevents uncontrolled cost growth. Enterprises should also create a 'use case driven' decision framework that defines when AI coding agents should be used and their appropriate levels of autonomy. Tasks can be classified into three models: developer-led, developer-with-agent, and fully agent-led.
Selecting models based on task complexity is another cost-saving tactic. Work should be broken into smaller tasks that can be performed by smaller models, with escalation only when complexity demands it. Engineering teams should route simpler, high-frequency tasks to smaller models and use frontier models only for complex and high-value work. Mandating specific context engineering practices can also help, such as training developers to optimize input context by including only relevant information, summarizing content, and eliminating unnecessary data.
Regular token usage reviews should be embedded into development cycles to identify inefficiencies and refine practices. 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. Leaders should not treat escalating AI coding costs as a reason to abandon AI or shift entirely to open models; instead, they should focus on optimizing costs without compromising value.
Starting small and focusing on context engineering first is recommended. Assessing current software engineering maturity and selecting the appropriate level of agent autonomy can yield up to 20% productivity gains, which is significant. For developers, mastering context engineering is not only beneficial for their employers but also for their career growth. The industry must recognize that token consumption requires active management to avoid cost overruns that could undermine the benefits of AI.
Source:InfoWorld News