KPMG: employee agent use doubled while enterprise rollout flatlined
KPMG's Q2 2026 AI Pulse survey reveals a sharp divide in how organizations are deploying AI agents: adoption among individual employees is accelerating rapidly, while broader enterprise rollouts remain constrained by cost concerns and unproven business returns.
The research, fielded between April 28 and May 25, 2026, across 2,145 senior leaders in 20 countries, found that adoption nearly doubled in a single quarter, with employee-level agent usage rising sharply. At the same time, enterprise-wide deployments have plateaued, suggesting that AI agents remain concentrated among early adopters within organizations rather than spreading across entire workforces.
The Adoption Paradox
The divergence points to a critical tension in the agent economy: enthusiasm among individual users is not translating into scaled organizational commitment. Three-quarters of leaders say AI is delivering meaningful value, yet nearly half of all organizations have scaled back, narrowed, delayed, or paused AI agent deployments because costs outweighed the returns. This friction suggests that while employees are experimenting with agents at a quickening pace, CFOs and enterprise stakeholders remain unconvinced that the productivity gains justify the infrastructure, licensing, and operational expenses.
What's Driving the Slowdown
KPMG's research indicates that the 49 percent pullback is concentrated among organizations attempting broader rollouts. Teams using agents for specific, high-value tasks—customer service, data analysis, code generation—are seeing results and continuing investment. But when companies attempt to deploy agents across multiple departments and functions simultaneously, the unit economics break down. License costs, integration work, training, and the overhead of managing agent failures accumulate faster than the measurable return.
The Real Story
For the agent economy, this data signals a maturation phase. The era of "agents everywhere" hype is giving way to a more disciplined market in which only agents solving demonstrable problems survive. Employee experimentation will likely continue—teams have already built workflows around agent tools, and switching costs are low. But until vendor pricing models stabilize and enterprises develop clearer frameworks for measuring agent ROI, enterprise adoption will remain selective and constrained.