By 2028, the expenses associated with AI coding may surpass those of developer salaries.

      According to Gartner, by 2028, the costs associated with the AI tools utilized by developers may exceed the salaries of the developers themselves. The expenses related to AI coding are rapidly increasing, and most companies are unaware of their spending habits. The surge in AI coding expenses is becoming substantial, with Gartner predicting that by 2028, these costs will surpass the typical developer salary. This increase is primarily due to the fact that each action an AI agent performs consumes tokens, leading to continuous costs.

      Tokens refer to the data units processed by an AI model, and under the current pricing models, a higher token count translates to a larger bill. Nitish Tyagi, a senior principal analyst at Gartner, remarked that companies are swiftly transitioning from experimenting with AI coding agents to large-scale deployment, though many are miscalculating the financial implications.

      This warning comes at an oddly favorable moment since AI coding tools are immensely popular. Engineers appreciate their functionality, and managers recognize improvements in speed. However, these tools are now poised to become more expensive than the labor they support. Gartner's message is clear: both popularity and costs are on the rise.

      The increase is already noticeable, with AI coding expenses skyrocketing from around $20 or $100 monthly per developer to between $2,000 and $5,000, as Tyagi informed The Register. This surge stems from a subtle shift in the pricing model; developers now pay based on consumption rather than a flat fee. An agent can utilize substantial tokens, causing an unpredictable billing situation. This dynamic has already caused some enterprise AI expenses to triple, despite declining token prices.

      Consumption-based pricing favors vendors as usage increases. The more an agent engages in writing, testing, and redoing tasks, the higher the costs. A single automated task can consume tokens that the developer isn’t aware of, and when multiplied across a team, the monthly bill escalates.

      This situation doesn’t indicate that the tools are ineffective. When used properly, they enable faster feature delivery and allow engineers to avoid mundane tasks. The concern lies in the disparity between anticipated savings and actual bills. At present, very few teams track this gap effectively, and even fewer act upon their findings.

      The core issue is a lack of transparency. Many vendors do not clarify how they bill or calculate token usage, meaning companies struggle to forecast costs and often deplete their budgets prematurely. Tyagi noted that most organizations are still lacking the frameworks to accurately measure costs in relation to business outcomes.

      Gartner details the implications: engineering leaders find it increasingly difficult to justify expenses linked to tokens. Budgets dwindle sooner than expected, and without a clear connection between expenditure and business value, budget discussions become awkward.

      Developers aren't the ones monitoring costs; their focus is on speed rather than expense. As Tyagi expressed, "token discipline" won’t emerge from developer choices alone. Without established regulations, costs may escalate faster than the productivity promised by the tools.

      Several factors are driving up expenses. Allowing agents to operate independently depletes tokens quickly. Excessive context windows—where a tool processes unnecessary information—are a contributing factor, and teams seldom create mechanisms to limit waste.

      The vendors of these tools have not implemented effective cost-control features, placing the responsibility for restraint on buyers who are largely unprepared. Additionally, as users become more comfortable with the tools, light users transition to heavy users. Gartner anticipates that model prices will increase as AI companies seek profits, leading to heightened usage at inflated costs.

      This is altering industry behavior, with some companies imposing limits on AI usage among employees. The most AI-intensive businesses are currently spending around $7,500 per employee monthly.

      The situation has spurred a market response. Database vendors are now marketing themselves as solutions to overspending on AI, claiming they can reduce the number of calls made by coding agents. There are also calls for an industry standards body to clarify billing practices.

      Even major companies are scaling back. Microsoft has quietly reduced its reliance on heavy Claude Code use due to costs, and GitHub has halted some Copilot sign-ups as demand for agents has impacted economics. While the tools are effective, managing the costs associated with using them at scale poses challenges.

      Gartner sees the broader market entering a new phase of growth and competition, which should eventually lead to improved cost management tools and clearer pricing structures. Currently, buyers are outpacing product capabilities, rapidly adopting tools that were not originally designed to be economical.

      Gartner’s recommendations emphasize maintaining discipline rather than withdrawal. It advises engineering leaders to categorize tasks into three categories: developer-led, developer-assisted, and fully agent-driven, with appropriate levels of autonomy for each.

      The first step should be effective model routing. A simple task does not require a sophisticated model. Gartner suggests that teams should direct frequent, uncomplicated tasks to smaller, less expensive models, reserving the costly ones for more complex challenges. Successfully implementing this strategy can lead to significant savings.

      Another important strategy is context engineering. Every additional