The transition from tentative pilot projects to the scaling of autonomous AI systems is accelerating across enterprise IT. According to the Global AI Pulse Q3 2026 report by KPMG International, based on a survey of 2,131 executives across 20 countries, average planned enterprise AI investments for the coming twelve months climbed to 210 million US dollars per company. This represents a marked rise from the 186 million US dollars recorded in the first quarter. Over the same timeframe, the share of organizations reporting substantial workforce adoption of autonomous AI agents jumped from 25 percent to 34 percent.
Despite this aggressive capital deployment, substantial blind spots remain regarding economic returns. While 89 percent of respondents in the KPMG survey report deriving measurable value from AI implementations, only 12 percent consistently track this return against the total cost of ownership. In response to rising operational complexity, enterprises are introducing formal governance layers. Approximately 55 percent of organizations have established a dedicated AI management framework, and 53 percent have anchored legal liability and decision accountability directly at the board or C-suite level.
The pressing need for governance is reinforced by findings from Redwood Software and Leger Opinion, who surveyed 300 IT, risk, and AI leaders. While 89 percent of respondents expanded their use of AI agents over the past year, with 54 percent scaling or running mature deployments, operational risks have materialized rapidly. A striking 77 percent of organizations experienced negative fallout from rogue or unmonitored agents. The most common issues cited were unpredicted cost explosions tied to cloud and API consumption, followed by operational downtime and compliance breaches.
Beyond monitoring failures, architectural technical debt continues to hinder corporate ambitions. A substantial 82 percent of respondents in the Redwood Software study identified legacy systems as a steeper barrier to enterprise AI rollouts than the absence of a high-level strategy. Currently, only 42 percent of surveyed organizations claim the capability to stably connect autonomous agents to core enterprise resource planning platforms and legacy backbones. Lacking deep integration, many agents operate in silos, requiring manual supervision and fragmented data retrieval.
Collectively, the data illustrates an industry wrestling with the operational friction of enterprise scale. The era of low-stakes experimentation is over, replaced by an urgent confrontation with infrastructure bottlenecks and unpredictable runtime expenses. Unless organizations implement rigorous cost caps and integrate autonomous agents directly into central enterprise architectures, escalating overhead risks erasing the operational efficiencies these tools were deployed to achieve.