Agentic AI Costs Spiral as Token Consumption Outpaces Enterprise Budgets
OpenAI has launched ChatGPT Work, a GPT-5.6-powered autonomous agent that integrates with tools like Google Drive, Slack, and Outlook to handle complex multi-step tasks independently. However, a 2026 Stanford Digital Economy Lab study found that agentic AI tasks consume roughly 1,000 times more tokens than standard chat interactions, with costs varying up to 30-fold between runs. Real-world cases illustrate the financial risk: Uber reportedly exhausted its entire 2026 AI budget in four months after deploying Claude Code, while Replit's gross margin swung from 36% to negative 14% due to agent-driven compute costs. Gartner projects that 40% of agentic AI projects will be cancelled by end of 2027, largely due to unclear ROI and uncontrolled unit costs. Industry analysts and practitioners argue that outcome-based pricing models — tied to tasks completed rather than API calls — along with task-level cost attribution and governance frameworks, are essential for enterprises to deploy AI agents sustainably.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in