Nearly half of large organizations have narrowed, delayed, or paused their AI agent deployments after operating costs began to exceed the value the systems produced, according to survey findings published by KPMG and circulating widely in coverage on August 8, 2026. The findings land as a reality check on a year of aggressive corporate enthusiasm for autonomous AI agents.

For executives and analysts weighing the economics of enterprise AI, AI Buzz Wire is tracking the gap between agent hype and measured returns with ongoing coverage of breaking AI news.

The cost-value gap

According to KPMG's findings, around 49% of large organizations have scaled back their AI agent rollouts because running the systems cost more than the value they generated. Only a small share of organizations reported reaching an established return on investment from their agent deployments. The result reframes one of the loudest narratives in enterprise technology: while AI agents are being embedded into more products than ever, turning that activity into positive economics remains elusive for many buyers.

The figure reflects a tension that has been building as companies move AI agents from pilots into production. Each agentic task — where a model reasons, calls tools, and takes multi-step actions — consumes significantly more compute than a simple chat prompt, because it often chains many model calls together. At scale, those costs compound quickly, and many organizations discovered the bill only after deployments were underway.

KPMG's survey detail

The findings align with KPMG's Quarterly AI Pulse Survey, which captured the perspectives of C-suite and business leaders at large organizations. In its U.S. edition, the survey gathered responses between April 28 and May 25, 2026 from 204 U.S.-based senior leaders representing organizations with annual revenue of $1 billion or more.

The survey identified understanding and managing the cost of operating AI at scale as an emerging challenge. While many companies have established foundational governance elements such as monitoring dashboards and approval processes, most still lack full, end-to-end visibility into AI-related costs in real time — a gap that becomes critical as organizations shift from isolated use cases to coordinated, enterprise-wide agent deployments.

Orchestration over standalone agents

Despite the pullback in rollouts, the broader direction of travel is not retreat but restructuring. According to the KPMG survey, organizations are advancing how agents are used, shifting from standalone applications toward more coordinated, enterprise-level orchestration. The growth in multi-agent workflows signals a move toward connecting activities across teams, systems, and decision points, using agents to align shared goals and automate work that spans functions.

In other words, companies are not abandoning AI agents; they are becoming more selective about where and how they deploy them, favoring integrated, high-value workflows over scattered experiments that are hard to justify on cost.

Humans still in the loop

The survey also found that the top risk-mitigation approach to AI agents over the next 6 to 12 months is a "human-in-the-loop" model, in which a person validates outputs but does not oversee every individual agentic action or decision. That preference reflects a practical compromise: organizations want the efficiency gains of automation but are not yet willing to let agents operate without a checkpoint on their results.

Employee experience emerged as another pressure point. The survey noted that employee concerns about working with AI agents are shifting, with many increasingly feeling the strain of added complexity and workload. Successfully scaling agent adoption, KPMG concluded, will require a sharper focus on employee experience, clearer alignment to outcomes, and more intentional approaches to driving engagement.

What it means for the market

Taken together, the findings describe a market entering a more disciplined phase. The first wave of AI agent adoption was driven by urgency and the fear of falling behind. The next wave, the data suggests, will be governed by unit economics: which agent workflows actually pay for themselves, how costs are tracked in real time, and where humans need to stay involved to keep outputs trustworthy.

For vendors, that means the bar is rising. Agent products will increasingly need to demonstrate measurable cost efficiency, not just impressive demos. For buyers, KPMG's data offers a useful benchmark: if roughly half of large organizations are pulling back because costs outran value, then cost discipline — not capability — may be the decisive factor in which AI agent deployments survive.

Stay Ahead of AI

The agent hype cycle is maturing into a cost-and-value reckoning. Follow the latest enterprise AI and agent analysis as the economics come into focus.

Explore AI Buzz Wire →