Decoding the 18-Fold Cost Increase
The headline-grabbing '18-fold' figure, unveiled at the Gartner IT Symposium/Xpo in Kochi, India, points to a troubling disconnect between the hype around AI and its actual financial payoff for businesses. While a majority of companies report seeing some
positive value from their AI projects, the investment needed to get there has exploded. According to Gartner, this surge is driven by a combination of rising infrastructure and integration expenses, the escalating cost of consuming AI models, and most importantly, the hidden price of managing the risks that come with more powerful AI. This isn't a signal to abandon AI, but a crucial call for companies, particularly in India, to be more selective and strategic, focusing on projects with clear, measurable business value.
Enter the AI Agent: More Power, More Problems
A key reason for this cost escalation is the rapid shift from simple AI tools to 'AI agents'. Unlike a chatbot that answers one question and stops, an AI agent is an autonomous system designed to achieve a goal. It can plan and execute multi-step tasks, access enterprise data, use other software tools, and adapt its actions without constant human direction. For example, an agent could be tasked to identify overdue invoices, draft and send reminder emails, and escalate non-responsive accounts to a human manager, all on its own. While incredibly powerful, this autonomy introduces significant risks if left unmanaged. Gartner found that while nearly three-quarters of organisations plan to deploy agentic AI soon, a mere 14% of CIOs in India and the Asia-Pacific region have the necessary governance capabilities in place.
What Is AI Agent Governance, Really?
Governance is the crucial framework of rules and controls that prevents autonomous AI from causing financial, operational, or reputational damage. It’s not a dusty policy document; it's a set of active, real-time guardrails built into your systems. Key principles include giving every AI agent a unique, trackable identity with a clear human owner who is accountable for its actions. It means enforcing 'least-privilege access', ensuring an agent can only access the specific data and tools needed for its immediate task, and nothing more. Crucially, it involves keeping a 'human in the loop' to approve high-impact or irreversible decisions and maintaining detailed audit logs to understand every action an agent takes. Without these controls, an agent could make a costly error in seconds, a mistake a human team would have had hours or days to prevent.
The Real Dangers of an Ungoverned AI
When an AI agent operates without strong governance, the consequences can be severe. Gartner predicts that by 2029, a majority of organisations running agentic AI will suffer a material service, security, or cost incident. Imagine an autonomous agent designed to optimise cloud spending misinterpreting a policy and shutting down a critical customer-facing service, leading to immediate revenue loss and reputational harm. Or consider a procurement agent with overly broad permissions being tricked into approving fraudulent payments. These are not far-fetched scenarios; they are the logical outcomes of deploying autonomous systems without an 'agent control plane'—a technical layer that can physically stop an agent from making a destructive error, something a written corporate policy can never do.
A Path Forward for Indian Businesses
For Indian enterprises eager to harness AI's potential, Gartner's warning is not a stop sign, but a roadmap. The goal is not to eliminate all friction, but to build in the right safeguards. The first step is to shift the focus from rapid experimentation to demonstrating measurable returns. Leaders must establish a clear framework for deciding which tasks are suitable for agents and what level of autonomy is appropriate. This means building governance capabilities in parallel with AI deployment, not as an afterthought. Businesses should invest in identity management for agents, enforce strict access controls, and build automated 'circuit breakers' that can pause or block an agent's actions when something goes wrong. Success with AI will no longer be about who adopts it fastest, but who controls it best.
















