Gartner Predicts AI Coding Costs Will Exceed Developer Salaries by 2028

On June 24, 2026, Gartner published a forecast that should change how every engineering organization budgets for AI. The research firm predicted that AI coding token costs will overtake the average developer's monthly salary by 2028. Hours earlier, a fintech startup employee in San Francisco had already proved the point by burning $81,267 in AI tokens in a single week, according to TechTimes.

The Slash Incident

Nicolas Brilliante, head of strategic verticals at Slash, a fintech startup valued at $1.4 billion, generated the $81,267 bill while using Anthropic's Claude to build a first-person browser game populated with internet meme characters. The project was part of a companywide initiative encouraging employees to embrace vibe coding, the practice of using AI to generate software through natural language prompts without formal architectural review, American Bazaar reported.

Slash posted the incident on X with characteristic startup humor, asking people to play the resulting game so the company could reclassify the expense as a marketing line item. Prediction market platform Polymarket amplified the post. Brilliante reposted with a note of disbelief, writing that he might become "a case study for how AI spend can get out of control."

Slash has since rolled back its companywide vibe coding initiative, TechTimes reported.

The Enterprise Pattern

The Slash incident would be a colorful one-off if it were not part of a broader pattern now visible across some of the largest engineering organizations in the world.

Uber deployed Claude Code to approximately 5,000 engineers in late 2025. By April 2026, the company had exhausted its entire annual AI coding tools budget in roughly four months. Monthly per-engineer costs ranged from $150 to $2,000, with Uber's CTO reportedly spending $1,200 in a single two-hour internal demo session, Forbes reported.

Microsoft's Experiences and Devices division, the team responsible for Windows, Microsoft 365, Outlook, Teams, and Surface, gave thousands of engineers access to Claude Code in December 2025. The tool proved "perhaps a little too popular," according to The Verge's reporting. Token-based billing consumed the annual budget ahead of schedule, and Microsoft directed engineers to transition to GitHub Copilot CLI by June 30, 2026, TheStreet reported.

Even Nvidia, whose core business is selling the hardware that AI runs on, acknowledged the cost challenge. Vice President of Applied Deep Learning Bryan Catanzaro told Axios that for his team, "the cost of computing power far exceeds the cost of employees."

Why Token Billing Breaks Enterprise Budgets

The core problem is structural. Traditional software licenses cost the same regardless of usage intensity. Token-based AI billing scales with consumption, and AI coding workflows are inherently iterative.

A session that begins with 50,000 tokens of context may be operating at 500,000 tokens after several hours of iteration. Each new instruction inherits the entire conversation history. By message 20, the context window contains every file read, every code block generated, every correction, and every revision. The engineer is paying for the same context information dozens of times over, often without realizing it.

The FinOps Foundation's sixth annual State of FinOps survey found that 98 percent of practitioners now actively manage AI spend, up from 63 percent in 2025 and 31 percent just two years earlier, People Matters SEA reported. The speed of the shift reflects the speed of the problem.

What to Watch

The Gartner forecast puts a timeline on what the Slash, Uber, and Microsoft incidents have already demonstrated. By 2028, the analyst firm projects that AI coding costs will match or exceed a developer's salary, making cost governance as important as code quality in AI-powered development workflows.

The immediate question for engineering leaders is not whether to use AI coding tools but how to budget for them. Starting a new session when switching tasks, narrowing the files passed to the model, closing unused tool integrations, and routing simple tasks to cheaper models are all practical ways to limit cost without sacrificing output quality. The companies that master AI cost discipline first will have a structural advantage over those that discover the problem when the invoice arrives.