When Prashanthi Kolluru, founder of KloudPortal, looks at how enterprises adopt artificial intelligence, she sees a pattern that worries her. Building AI agents is no longer the hard part. Operating them is. In a Council Post published on Aug 10, 2026, at 06:45am EDT, Kolluru argues that CIOs and technology leaders are more concerned about rising AI operating costs than model selection or prompt engineering. The article, published under Forbes Councils, reflects the opinions of the author as a member of the Forbes Technology Council, an invitation-only community for CIOs, CTOs, and technology executives.
The Shift From Building to Operating
The past year brought a fundamental change in how companies approach AI. Building agents became easier, but operating them became the challenge. Kolluru, whose company KloudPortal helps global capability centers (GCCs) hire product-ready engineering pods, draws on experience with enterprise clients to make her case. Most companies planned for development costs, but few anticipated the ongoing management costs for dozens or hundreds of agents.
AI agents do not sit idle after deployment. They operate continuously. They retrieve data, evaluate prompts, invoke models, connect to APIs, apply business rules, log activity, and route sensitive responses through human review. Each step is inexpensive in isolation, but collectively, these actions create a different operating model than traditional software. Enterprise software projects typically have predictable cost patterns, with the majority of spending during development. AI agents run on a consumption model where every interaction carries operational cost.
McKinsey research shows that enterprise AI adoption continues to accelerate across multiple business functions. Yet higher adoption does not automatically improve economics. Often the opposite happens. As business teams gain confidence, they request more. They want richer responses, deeper integrations, more internal knowledge access, and additional monitoring. These requests lead to more model calls, larger context windows, more retrieval operations, and heavier governance requirements.
Why Costs Climb After Launch
Organizations underestimate the trade-off because they evaluate AI like conventional software. That is a mistake. AI runs on a consumption model, and every interaction adds to the bill. Technology budgets often omit costs for monitoring output quality, evaluating model performance, updating prompts, maintaining data connections, enforcing governance, and measuring outcomes. As the agent count grows, this work becomes daily operations, not a one-time project.
The gap between expectation and reality is stark. McKinsey's November 2025 State of AI survey found that only 23% of organizations have scaled agentic AI in at least one business function. Meanwhile, 39% are still in the experimentation phase. Fewer than 10% of respondents report scaled deployment in any single function. These numbers suggest that most companies are still trying to figure out how to move beyond pilots, and the operational burden is a major reason why.
Deloitte's State of AI in the Enterprise research adds another layer. Only 1 in 5 companies has a mature model for governing autonomous AI agents. Governance, risk management, and operational readiness are among the biggest hurdles to scaling AI beyond pilots. Without proper governance, costs spiral and risks multiply. The research from both firms points to the same conclusion: the operational side of AI is where enterprises struggle most.
Rethinking How to Measure AI Value
Kolluru argues that cost per token is too narrow a lens for evaluating AI spending. The better question is whether each agent delivers sufficient business value. An agent automating a high-volume customer process can outweigh several cheaper agents solving narrow problems. Leaders should ask a specific question, according to Kolluru. "What business outcome does this agent improve, and what does it cost to deliver that outcome reliably?"
That question shifts the focus from unit costs to outcomes. It forces organizations to think about the value an agent creates, not just the price of each model call. A single expensive agent that handles thousands of customer interactions may be far more valuable than a cluster of cheap agents that barely move the needle. The measurement framework matters as much as the technology itself.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
This is a leadership challenge, not just a technical one. Managing a growing portfolio of agents with clear operating costs, governance, and business impact requires discipline. It requires asking hard questions about which agents earn their keep and which ones quietly drain resources. The organizations that get this right will be the ones that scale AI with confidence.
FinOps for the AI Era
Cloud computing gave rise to FinOps, a practice for managing infrastructure spending with visibility and accountability. Enterprise AI is creating the same need, just faster. Organizations are beginning to apply FinOps principles to AI for cost visibility and accountability. The idea is straightforward: treat AI operations like any other significant expense, with clear ownership and regular review.
The best-positioned organizations build operational discipline. They review usage patterns, measure outcomes, retire redundant agents, and treat AI operations as an ongoing capability. Cost cannot be eliminated, but it can be made explainable. That explainability is what allows leaders to make informed decisions about where to invest and where to cut.
Enterprise AI is moving beyond experimentation into day-to-day operations. This transition reveals that budgeting assumptions are different from actual costs. The gap between what companies planned to spend and what they actually spend on AI operations is a recurring theme in Kolluru's analysis. It is a gap that catches many organizations off guard.
The Leadership Imperative
The article makes clear that this is not a problem that technology alone can solve. Managing a growing portfolio of agents with clear operating costs, governance, and business impact is a leadership challenge. It requires a different mindset from the one that drove initial AI adoption. Building agents was the easy part. Running them well is the hard part.
Organizations that see the shift early will scale AI with confidence. Those that ignore the operational costs will find their AI initiatives stalling. The evidence from McKinsey and Deloitte suggests that most companies are still in the early stages of this journey. The 23% that have scaled agentic AI in at least one business function are the exceptions, not the rule.
Kolluru's message is practical. She does not argue that AI costs can be avoided. She argues that they can be managed. The tools and practices exist. FinOps principles can be applied to AI. Operational discipline can be built. The question is whether organizations will do the work.
The hidden costs of AI agents are not hidden to those who look. They are visible in every model call, every retrieval operation, and every governance review. The organizations that acknowledge these costs and build systems to manage them will be the ones that succeed. The ones that pretend AI operates like traditional software will struggle.
The article concludes with a clear call to action. Enterprise AI is moving beyond experimentation into day-to-day operations, revealing budgeting assumptions that are different from actual costs. Managing a growing portfolio of agents with clear operating costs, governance, and business impact is a leadership challenge. Organizations that see the shift early will scale AI with confidence.