What's Happening?
A recent IDC survey, commissioned by DataRobot and published in December 2025, found that 92% of organizations utilizing agentic AI reported that their costs were higher than anticipated. The survey, which included 318 senior decision-makers from organizations with
over 1,000 employees across the United States, United Kingdom, and Ireland, also indicated that 71% of respondents felt they had little to no control over these escalating costs. This issue stems from the structural nature of agentic workflows, where costs scale not by user count, but by the number of steps a task takes, which is not fixed. Agents can retry failed tool calls, loop while waiting for slow systems, or chain multiple calls, leading to unpredictable and often higher expenses that may only become apparent when the invoice arrives.
Why It's Important?
The findings highlight a significant challenge for U.S. businesses investing in agentic AI: a lack of cost predictability and control. This can lead to budget overruns and hinder the widespread adoption and effective scaling of AI technologies. Companies that cannot accurately forecast or manage AI-related expenditures may face financial instability, reduced ROI on AI investments, and difficulty in justifying further AI initiatives. The problem affects various industries, as agentic AI is increasingly integrated into operations. Without clear cost governance, businesses risk not only financial penalties but also a slowdown in innovation as they become wary of deploying advanced AI solutions due to unforeseen expenses. This issue underscores the need for robust financial planning and transparency in AI deployment strategies.
What's Next?
To address the issue of unpredictable agentic AI costs, the survey suggests implementing cost ceilings at the design stage of AI features. This involves setting a maximum token or dollar spend per task before an agentic feature is deployed. Organizations need to decide in advance what actions the agent should take if it approaches this limit, such as stopping, escalating to a human, or reverting to a more cost-effective path. Furthermore, it is crucial for every agentic workflow to have a clear owner who is alerted when cost ceilings are hit, similar to how engineers are paged for service outages. This proactive approach aims to transform cost management from a reactive invoice review to a strategic design consideration, ensuring that AI deployments remain financially viable and sustainable.
Beyond the Headlines
The challenge of managing agentic AI costs extends beyond mere financial implications, touching upon ethical and operational dimensions. The lack of transparency in AI cost structures can erode trust in AI systems and their providers, as businesses struggle to understand and justify expenditures. This could lead to a broader skepticism towards advanced AI adoption, potentially slowing down technological progress and competitive advantage for U.S. companies. Moreover, the need for human intervention when cost ceilings are reached introduces a new layer of operational complexity and decision-making, blurring the lines between automated and human-driven processes. This situation necessitates a re-evaluation of how AI is integrated into business models, emphasizing not just performance but also economic accountability and transparent governance from inception.
