Technology & Gadgets | June 2026 | 4 min read
| Gartner warns AI coding costs may overtake average developer salaries by 2028 as token usage and agent workflows drive bills higher. |
⚡ Quick Takeaways
- Gartner expects AI coding costs to overtake the average developer salary by 2028.
- The warning is based on rising token consumption and a shift to usage-based pricing.
- The forecast uses a global average developer salary of about $2,000 per month.
- Agent-based coding workflows and large context windows can quickly increase token use.
- Gartner says companies need governance, model routing and context discipline to control costs.
AI coding tools are moving from helpful productivity add-ons to major software engineering expenses, with token-based bills becoming harder for companies to forecast and justify.
| 2028 Gartner cost warning | $2,000 global monthly salary baseline | $20,000+ reported extreme monthly usage |
Gartner Warns AI Coding Bills Could Catch Salaries
Companies could soon spend as much on a developer’s AI coding usage as they do on the developer’s salary, according to Gartner. The research firm expects AI coding costs to overtake the average developer salary by 2028 as token consumption grows and more vendors move from fixed subscriptions to consumption-based pricing.
The forecast does not mean every developer’s AI bill will exceed every salary. Gartner’s comparison is based on a global average developer salary of about $2,000 per month. In high-pay markets such as the United States, salaries can be far higher, so the comparison is best understood as a warning about average global cost pressure rather than a universal payroll replacement number.
Still, the warning is serious because some organisations are already seeing unusually large bills. Gartner analyst Nitish Tyagi has cited cases where one developer used around $20,000 worth of AI resources in a month, while a business user generated an even larger bill. These examples show how fast usage-based pricing can surprise companies when there are no strong controls.
| Important Context: The salary comparison uses a global average, so it should not be read as a claim about every developer in every country. |
Why Token-Based Pricing Is Hard to Control
AI coding tools are often priced around tokens, the units of text and code processed by large language models. The more context, prompts, generated code and agent steps a tool uses, the more tokens it consumes. Under a fixed subscription, companies could predict software spending more easily. Under consumption pricing, the bill can change sharply from one month to the next.
Gartner says many AI coding vendors still lack mature cost optimisation and transparent billing tools. That makes it difficult for engineering leaders to understand exactly which tasks, teams or workflows are driving spending. If usage rises faster than productivity, companies may struggle to prove that the cost is worth it.
Prices may also stay under pressure as AI vendors invest heavily in infrastructure and try to make their services profitable. Even if individual model costs fall in some areas, wider use, larger workflows and more automated agents can still push the total bill higher.
| Cost Driver | What Happens | Risk | Control Needed |
| Agent workflows | Tools act with limited human input | Token use keeps running | Set autonomy limits |
| Large context | Models process too much code or text | Unnecessary token spend | Use context engineering |
| Model choice | Frontier models used for simple tasks | Higher cost per request | Route to smaller models |
| Weak reporting | Teams cannot trace usage clearly | Budget overruns | Track token thresholds |
AI Agents Can Burn Through Budgets Faster
Agent-based coding tools create a different cost problem from older code assistants. A simple assistant may answer a prompt or suggest a function, but an agent can plan, search files, rewrite code, run checks and repeat steps. That autonomy can be useful, but it can also keep consuming tokens long after the original instruction was given.
Large context windows add another risk. When developers feed entire repositories, long logs or unnecessary documentation into a model, the tool may process far more information than the task requires. Gartner recommends context engineering, which means giving AI systems only the relevant information, summarising where possible and removing unnecessary data before the model runs.
| Cost Alert: Agentic workflows can be helpful, but without task limits and monitoring they can quietly consume large token budgets. |
Governance Will Decide Whether AI Coding Pays Off
Gartner says companies need a more disciplined operating model for AI coding. That includes deciding which tasks should be developer-led, which should use a developer with an agent, and which can safely be agent-led. Not every bug fix, refactor or test-writing task needs the most advanced model.
Engineering teams can also reduce costs by using smaller models for simpler tasks and reserving frontier models for complex, high-value work. Token thresholds, escalation policies and automated monitoring can help prevent uncontrolled spending. Regular token usage reviews during sprint retrospectives can also identify wasteful workflows before they become normal practice.
The deeper issue is return on investment. If AI tools help developers ship better software faster, some extra cost may be justified. But if bills rise without measurable improvements in delivery speed, quality or business value, AI coding tools could become another expensive software line item that finance teams push back on.
| Practical Step: Companies should match model size to task value, limit unnecessary context and review high-token workflows regularly. |
Disclaimer: This article is for general information only and is based on Gartner’s public forecast and verified technology reporting available at the time of writing. AI pricing, salaries and vendor policies may change quickly.

