The declining price of language model tokens is far from translating into savings for enterprises. On the contrary, Gartner warns that inference costs per agentic workflow may skyrocket more than five times by the end of 2028, even as foundational models become increasingly cheaper. The paradox of agentic AI economics lies in a deceptively simple mechanism: as models become more cost-efficient per token, users push the technology toward increasingly complex tasks, consuming vastly more tokens in the process and erasing any per-unit discount.
The finding, reported in August 2026 by Dan Robinson at The Register, synthesizes research from the analytics firm on what happens when organizations move beyond basic chatbots to build autonomous systems capable of planning, executing, and self-evaluating tasks. The cost difference is not marginal — it is structural. A single agentic workflow can consume dozens of times more tokens than a simple chatbot question, because the system must continuously reason, negotiate with other models, validate outputs, and iterate on itself.
Will Sommer, senior analyst at Gartner, frames the difference in practical terms. A chatbot reads a query, interprets intent, and produces a probabilistic answer. An agent, by contrast, must maintain a closed loop of reasoning: assess the current state, decide the next action, execute, verify the result, and if necessary restart the cycle. Each additional iteration multiplies inference costs. According to Gartner, routing a task to an agentic reasoning model increases inference costs at least fivefold, and potentially far more as task complexity grows.
This is not merely theory — companies are already feeling the impact. The Register reports that the shift by some AI providers from flat-rate subscription pricing to usage-based billing has exacerbated the problem. Token-heavy workflows can produce surprising invoices under the new model. The same report cites a previous The Register article published in July 2026, documenting how C-suite executives face incomprehensible AI bills after migrating to consumption-based pricing.
Gartner's context is consistent with earlier predictions from the same firm. In June 2026, the company stated that AI coding costs would surpass the average developer's salary by 2028, precisely due to rising token consumption and the adoption of consumption-based licensing models. In May, Gartner predicted that 40% of organizations would decommission or demote agentic AI projects due to cost and performance problems — a prediction that sounded pessimistic in May but gains contours of reality when considering the exponential consumption curve.
The irony is that the very factors making agentic AI so attractive are the same that make its financial operation unsustainable without rigorous governance. The ability of an agent to execute multi-step workflows with minimal human intervention is exactly what drives its disproportionate consumption of inference cycles. It is the same principle that explains why streaming users consume far more content when quality improves — the enhancement of experience drives demand expansion, not total cost reduction.
Companies seeking real return from agentic AI will need to adopt sophisticated inference optimization strategies, intelligent task routing, and model orchestration. This means, in practice, assigning each task to the most cost-efficient model capable of executing it, rather than simply scaling to the most powerful model. Gartner recommends this granular task-by-task selection approach but acknowledges that practical implementation is challenging: it requires real-time cost monitoring per workflow, capability mapping across models, and routing infrastructure that can be as complex as the agents being orchestrated.
NVIDIA, OpenAI, Google, and other technology giants continue to push agentic AI as the next frontier of digital transformation, with billion-dollar investments in inference infrastructure. But Gartner's signal is clear: without architectural change, agentic economics is an illusion. Tokens are getting cheaper per unit, but the total bill keeps rising — and the trend is accelerating.
Sources: The Register, Gartner, CXOToday
✓ Independent sources cross-checked and verified before publishing
