As businesses race to deploy autonomous AI agents, tech research firm Gartner is raising a yellow flag. The excitement over AI's potential is meeting a harsh reality: soaring costs and a dangerous gap in governance.
The 18-Fold Cost Shocker
The most startling finding from Gartner,
discussed at its IT Symposium in Kochi, is that the investment required to generate positive returns from AI has increased 18-fold in just two years. This figure highlights a growing disconnect between enthusiasm for AI and its actual commercial payoff. While early experiments might seem cheap, scaling AI across an organization introduces significant, often unforeseen, expenses. These include costs for technology, infrastructure, implementation, specialised talent, and the continuous oversight needed to manage these complex systems. According to Gartner analyst Neha Kumar, this shifting economic landscape means businesses must be far more selective about which AI projects they pursue.
Beyond Pilots: The Real Cost of AI
Why the dramatic cost increase? The move from small-scale pilots to full production is a primary driver. Gartner notes that organizations get a 'rude awakening' on costs when they transition from experimentation to enterprise-wide deployment. Production AI involves sustained use, which balloons expenses. Unlike a simple chatbot, an AI agent can trigger chains of actions from a single user request, potentially leading to tens or hundreds of Large Language Model (LLM) calls. This nonlinear consumption can cause costs to spiral if not properly governed. Furthermore, Gartner predicts that through 2028, the cost of 'inference'—the process of running a live AI model to generate outputs—will account for at least 70% of a model's total lifetime cost, far exceeding the initial training expense.
The Dangerous Governance Gap
Alongside rising costs is an even more pressing issue: a severe lack of governance. Gartner warns that many companies are rushing to adopt autonomous AI without the necessary guardrails. In a recent survey, while 71% of organizations plan to deploy AI agents within a year, only 21% have the governance capabilities to oversee them. This creates a massive operational readiness gap. Another Gartner survey found that 95% of organizations had already faced at least one AI agent-related incident in the past year, such as cascading API loops or cloud resource exhaustion. Despite this, a surprising 70% of infrastructure and operations (I&O) leaders believe their AI agents operate safely—a confidence Gartner suggests is misplaced.
Risk of 'Business-Critical' Failures
The consequences of this governance deficit are severe. Gartner predicts that by 2028, 40% of I&O organizations that scale AI without strong, dynamic governance will suffer a business-critical service disruption. The issue is that most existing IT risk tools are considered outdated for handling autonomous machine behaviour, with 57% of I&O leaders admitting their current governance is weak. Runaway agent errors, infinite logic loops, and unauthorized system modifications are real risks. To counter this, Gartner recommends a new, proactive approach to infrastructure governance that treats AI agents as 'nonhuman identities' requiring specific controls, permission models, and audit capabilities.
What Should Businesses Do Now?
The message from Gartner isn't to abandon AI, but to approach it with strategic discipline. Leaders are advised to become more selective about AI projects, focusing on those with a clear and justifiable return on investment. Establishing a central decision-making authority for AI strategy can help weigh value against risk consistently. On the technical side, it's crucial to track and manage the total cost of ownership, not just the initial subscription or token fees. Implementing real-time governance with 'circuit breakers' to halt runaway processes is critical for safety. Ultimately, business leaders must shift their focus from simply asking if AI can do a task to whether the productivity gains truly justify the full, long-term cost and operational risk.
















