AI Agent has become one of the hottest keywords in enterprise AI adoption over the past two years, yet most organizations have not crossed the line into letting it actually do the work. In November 2026, PYMNTS Intelligence published The Agentic Edge Index, based on a survey of 1,687 senior executives at large U.S. companies. The findings show that the question is no longer whether businesses use AI, but whether they are willing to let it take action.
From Reading to Taking Action: Where Are Enterprises Getting Stuck?
The index maps 75 enterprise activities across eight business functions and uses a five-level scale to measure the depth of AI involvement:
- Levels 1 to 3: Chatbots and generative AI help employees perform tasks, while humans remain in control.
- Levels 4 to 5: AI Agent takes over execution, with humans responsible only for approvals or handling exceptions.
The report calls the boundary between Levels 3 and 4 the “agent line.” Most enterprises remain below it.
The data also illustrates just how quickly AI adoption is advancing. At the end of 2025, only 7% of senior executives were using or testing AI Agent, while 67% of enterprises gave AI Agent no access at all. By 2026, however, 89% of enterprises were using AI Agent in at least one business function, and only 2% still refused to grant it any access.
What most enterprises have granted AI Agent is essentially read-only access: 78% allow it to read company systems but not take action; 18% permit AI Agent to complete tasks after human approval; and only 2% of all surveyed enterprises allow it to handle routine work independently. This explains why 90% of enterprises score below 25 on the 100-point index. Even the most advanced companies have yet to achieve independent execution at scale.
The research team created a separate maturity index because asking whether a company “uses AI” no longer reveals meaningful differences. In the 2026 survey, 97% of senior executives said AI was generating positive returns. Even among the lowest-ranked enterprises, 94% gave the same response. The question had therefore lost its ability to differentiate companies, making it necessary to ask instead: “How far has AI progressed?”
Looking at the timeline, large enterprises in 2024 were using an average of only six of the 13 tracked generative AI use cases, mostly for routine tasks such as drafting content and responding to customer inquiries. By early 2025, 46% of generative AI work could already be completed with little to no human intervention. The share of senior executives who described AI returns as “highly positive” also rose steadily, from 20% in early 2024 to 57% in 2026. Building on this foundation, AI Agent has been introduced into a growing range of work scenarios.
What Leaders Are Doing: The Gaps in Revenue Growth and Manual Work Replacement
When enterprises are divided into three groups based on their index scores, the share of top-tier companies achieving revenue growth above 15% is more than double that of the bottom tier. Meanwhile, 82% of top-tier enterprises are growing both revenue and profit margins, compared with only 66% of bottom-tier companies. For every 10-point increase in the index, an enterprise’s likelihood of achieving revenue growth above 15% more than doubles.
The key differentiator is how much manual work AI Agent has replaced. Eighty percent of top-tier enterprises say AI has already replaced some or most manual operations, compared with only 45% of bottom-tier companies. As index scores rise from 15 to 25, the share of enterprises reporting that AI has replaced tedious manual work jumps from 58% to 86%.
Top-tier enterprises are also more willing to invest. Over the past year, they spent an average of US$14.5 million on emerging AI applications, compared with US$9.1 million among bottom-tier enterprises. Twenty-two percent of top-tier companies plan to increase their AI budgets by at least 15% next year, versus only 8% of bottom-tier companies. Across the full sample, 79% of large enterprises plan to raise their AI budgets next year, and not a single company intends to reduce spending. The difference is simply that top-tier enterprises are investing more aggressively.
Top-tier companies are also more confident about returns: 64% believe their current AI returns are already “highly positive,” yet 66% also expect it will take more than three years to realize the full return. In other words, they are capturing short-term benefits while recognizing that AI is a long-term investment, not a one-off project.
Notably, top-tier enterprises are not spreading AI Agent evenly across every department simply to showcase its use. Instead, they prioritize the highest-risk, highest-impact activities: fraud and financial crime detection, compliance monitoring, credit risk, cybersecurity monitoring, treasury management, and month-end and year-end close processes. Average AI adoption across these high-risk activities has reached 44%, compared with 39% for general business activities, while cybersecurity monitoring is even higher at 55%. Top-tier enterprises use AI Agent in 45% of these high-risk activities, compared with only 26% among bottom-tier companies. Beyond risk management, top-tier companies are also more proactive in applying AI Agent to processes that directly affect revenue, such as contract and proposal generation (46% versus 19%) and invoice processing and reconciliation (46% versus 26%).
What This Means for Decision-Makers in Traditional Industries
The survey’s most practical finding is that more than half of senior executives attribute stalled progress to people, processes, and corporate culture. The technology itself is not the primary bottleneck. Model capabilities are no longer the issue. The real questions are who should approve an action, which steps can be delegated, and who is accountable when something goes wrong.
For companies in traditional industries that are still evaluating whether to adopt AI Agent, report author Karen Webster offers a highly practical starting point:
- Stop counting how many AI tools the company uses. Instead, ask what stage each task has reached and whether it is being performed by a person or by AI Agent.
- Start in areas where trust is already well established, such as risk management and cybersecurity teams that have used machine learning for years, and make them the proving ground.
- Apply AI Agent to work that directly affects revenue and cash flow, such as proposals, invoices, and incident response, rather than beginning with showcase projects such as customer service chatbots.
- Reserve part of the budget for people, not just software licenses. Redefining responsibilities and designing approval workflows often take more time than implementing the tools themselves.
The survey also highlights an easily overlooked issue: 84% of senior executives expect their businesses to grow over the next three years without a corresponding increase in headcount, while 75% expect roles and required skills to change. For enterprises adopting AI Agent, reallocating work and approval authority will be more important than simply adding or reducing staff.
In EgentWrX’s experience supporting enterprises through implementation, this is also the path most companies follow when introducing digital twins: first allowing the digital twin to read data and prepare drafts, then assigning it complete tasks once the workflow has stabilized and access boundaries have been clearly defined. The PYMNTS data confirms one point: enterprises willing to cross that line are already beginning to pull ahead in growth.
FAQ
Is the sample size of this survey reliable? The research is based on 28 surveys conducted by PYMNTS Intelligence among U.S. enterprises with annual revenue of more than US$1 billion, incorporating responses from a total of 1,687 senior executives. The index’s group comparisons are based on a 2026 sample of 366 AI decision-makers.
Why are index scores generally so low even though enterprises are already using AI? Because the index measures how much work AI performs and the level of autonomy it has achieved. Whether someone in the company uses AI is no longer the key issue. Most enterprises still rely on chatbots or generative AI as support tools. Very few allow AI Agent to execute tasks independently while humans handle only exceptions. As a result, scores remain low even though AI is used almost universally.
Are leading enterprises growing revenue faster because they use AI Agent? The report emphasizes that the relationship is correlational, not causal. Deeper AI adoption may help enterprises grow faster, but it is also possible that fast-growing companies are more willing to invest in AI. Both forces are likely at work simultaneously.
Is it too late for Taiwan’s traditional industries to catch up? The data suggests that the gap is only beginning to widen, and most large U.S. enterprises still remain below the agent line. Rather than chasing the latest technology, companies should begin by assigning AI Agent one stable, high-trust workflow, then establish approval mechanisms and clear accountability. This is a more practical starting point than attempting a company-wide rollout all at once.
References
- PYMNTS Intelligence, The Agentic Edge Index: How Leading Companies Put AI to Work, enterprise research report page
- PYMNTS, How Large US Companies Put AI Agents to Work, report press release
- Karen Webster, The Agentic Edge: Who Has It and How to Get It, an in-depth analysis by the CEO of PYMNTS
