Genpact and HFS Research report says 92% of senior executives believe agentic AI will fundamentally change how work is executed, yet most companies are still not set up to let these systems operate with broad autonomy. The study suggests the main barrier is no longer model capability; it is organizational readiness.
The report, Autonomy Requires Trust in AI, is based on a survey of 545 senior executives across 11 industries, plus interviews with leaders from Fortune 2000 companies. Genpact and HFS say the research was designed to identify what keeps enterprises from moving agentic AI from a productivity tool to an autonomous execution layer inside the business.
92% of respondents said agentic AI systems, meaning systems that can direct tasks and make decisions autonomously, will fundamentally change how work is done. But the same research also found that nearly 80% of organizations still run these systems in supervised mode, with humans keeping final approval on most actions.
[Also Read: 93% of Indian Businesses Expect AI ROI Within Three Years, Finds SAP’s Value of AI Report 2025 ]
That gap suggests to us that companies are willing to adopt agentic AI in principle but are not yet willing to hand over meaningful decision-making rights in practice. In enterprise terms, this is not a tooling problem. It is a control problem.
Only 22% of organizations said they are comfortable granting AI agents broad autonomy, and HFS frames accountability, not capability, as the real scaling barrier. That is consistent with a wider HFS theme appearing in its April 2026 research feed, where the firm says scaling autonomy requires structural redesign across the enterprise.
The report also says 33% of respondents see unprepared business processes as the top obstacle to agentic AI adoption. That is an important detail because it pushes the discussion away from model quality and toward workflow design. In other words, the bottleneck is often the process architecture around the AI system, not the AI system itself.
[Also Read: Nokia and AWS Unveil Agentic AI-Powered 5G Network Slicing With du and Orange ]
Spending is rising, but measurement is lagging
Despite the caution, the commercial direction is still upward, Genpact says enterprises expect investment in agentic AI to rise 38% in the next year, and that companies expect to scale these systems within an average of 17 months. At the same time, 67% still rely on productivity metrics that do not properly measure the value of independent decision-making.
Budgets move first, but management systems move later, when new technology gets funded before new performance measures are established, organizations often end up over-reporting activity and under-reporting outcome. This report suggests agentic AI is entering that phase now.
What changes inside the enterprise
The report points to a structural change in how companies may organize work, 44% of executives expect fewer management layers as agentic AI absorbs coordination tasks. The skills in demand are also changing, with workflow orchestration and integration, data engineering, and monitoring and observability emerging as the most needed capabilities.
The skill mix is moving away from building isolated AI use cases and toward running AI inside business operations. The question is whether the organization can manage exception handling, escalation processes, accountability, and process redesign at the same time.
Genpact and HFS are making a wider point: agentic AI will not scale because executives are excited about it. But it will scale only when companies redesign accountability, measurement, people, and process together.
[Also Read: Kyndryl unveils agentic service management for AI-native infrastructure ]
That distinction matters for readers tracking the agentic AI market, the next competitive advantage will likely not belong to the company with the most AI pilots. It will belong to the company that can safely convert supervised automation into governed execution across real workflows. That is a harder task, but it is also the one that determines whether agentic AI becomes operational infrastructure or stays a controlled experiment. This conclusion is an inference from the report’s findings on supervision, process readiness, and accountability.
