Agentic artificial intelligence (AI) is no longer just an idea or buzzword. Now, for the first time, enterprises are starting to put the technology into practice.
Agentic AI autonomously takes actions and makes decisions independent of human guidance beyond a set of predefined inputs. The software goes beyond the content-creating capacities of generative AI (gen AI), which relies heavily on human control. Today, more than one in 10 U.S. enterprises have adopted or are testing agentic iterations. This leap forward marks a striking shift from just this past spring, when none were doing so.
Firms are looking to both in-house tech teams and external partners and vendors to get their agentic AI efforts running. Enterprise chief financial officers (CFOs) are enthusiastic about using the technology for strategic planning and financial reporting. But they remain cautious about applying it in complex and high-stakes areas such as treasury management. Ultimately, the technology will gain fuller, more widespread use only if CFOs are willing to grant it access to internal directories of clients and vendors and allow it to autonomously initiate actions, like payments—and most aren’t ready to do that just yet.
These are just some of the findings in this edition of the CAIO Report, “How Agentic AI Went From Zero to CFO Test Runs in 90 Days,” a PYMNTS Intelligence exclusive. This edition examines the rise of agentic AI at U.S.-based enterprises. It draws on insights from a survey of 60 CFOs working at U.S. firms that generated at least $1 billion in revenues last year. The survey was conducted from July 15, 2025, to July 25, 2025.
Agentic AI is moving from concept to reality.
For more than a year a heavily hyped buzzword, the cutting-edge technology is quickly gaining a toehold in corporate operations. As of July, 6.7% of U.S. enterprise CFOs are using agentic AI, and another 5.0% are piloting or testing it. An additional 8.3% are exploring adoption in the next 12 months. All told, 35% of enterprises are already using or considering adopting agentic AI at some point.
This marks a dramatic change from just this past spring. As recently as May, only 1.7% said they were exploring the possibility of adopting the technology within the next year. At that time, none were using or piloting it in test trials.
Then in July, 65% of CFOs reported no plans to adopt agentic AI. This was down sharply from 85% in May—a steep drop in skepticism and a critical shift toward adoption.
Perhaps unsurprisingly, the industry that invented the software—tech—has been the quickest to adopt the technology. Conversely, services firms, whose workflows depend heavily on data privacy and security that if breached can cause cascading disruption to operations, brand and image, have been hanging back. What’s clear is that companies that successfully use gen AI are creating organizational buy-in for agentic. Businesses that have seen the most positive return on investment (ROI) from gen AI are also proving the quickest to adopt its agentic offshoot.
The applications in which firms use gen AI also affect their use of agentic AI.
Enterprises that have used gen AI for the most high-impact applications—those carrying significant business value and risk, such as sending payments—are far more likely to already be exploring agentic AI. In fact, one in four enterprises utilizing gen AI in high-impact functions and processes are currently using or piloting agentic. Half of these firms are at least considering adopting the technology at some point. It seems that these enterprises trust agentic AI more after seeing positive early outcomes from less advanced iterations of AI. These businesses, especially tech companies, some of which have invented AI technologies, may also have a stronger appetite for experimentation.
By contrast, enterprises using gen AI for low-impact tasks such as summarizing emails, helping employees find information and generating summaries, are holding off on agentic AI. No enterprises are currently using or testing it. These firms may be reluctant to take on the risks of new technologies and applications due to limited evidence that it produces proven returns, resource constraints or organizational aversion to risk.
From in-house development to external partnership, enterprises are approaching agentic AI from all angles.
Enterprises that are adopting or considering agentic AI are trying a variety of approaches. The majority of these firms—52%—are building in-house capabilities with internal AI/engineering teams. The same share is partnering with external AI vendors such as OpenAI, Google, Anthropic, Salesforce, SAP and FinTechs or with consultants.
Among firms that are already using or piloting agentic AI, 71% are building in house, and 43% are working with external partners. Additionally, 43% are also procuring off-the-shelf agentic AI tools. For now, enterprises are trying a range of different methods to see what works best. This signals that the market for vendors and other providers is fragmented and highly competitive.
It is getting easier for firms to develop and train agentic models internally. In-house development may enable greater long-term control, customization and deep organizational expertise. Building in-house expertise might take longer and require bigger investments, but it might help a company customize and control what it aims to achieve. By contrast, external partnerships and off-the-shelf tools may enable firms to move quickly with lower upfront investment. The range of approaches and solutions underscores how two selling points—control and speed—are jostling for dominance and market share.
Enterprise CFOs are interested in using agentic AI for strategic planning, but cautious about using it for treasury, risk and compliance.
Companies in the education industry say they rely on agentic AI to automate the preparation of financial reports for board reviews and strategic meetings. Food and beverage distributors use the software to detect anomalies and discrepancies in financial records. In healthcare, agents reconcile accounts payable and receivable transactions in real time. Real estate companies use the technology for dynamic forecasts that update in real time, keeping financial projections aligned with current market trends. Technology companies use it to identify overspending relative to competitors or industry benchmarks. In the travel and transportation industry, enterprises are deploying it to enhance risk modeling for volatility in fuel prices and currency exchange rates. A company’s industry defines its use cases.
That said, seven in 10 enterprise CFOs report being very or extremely interested in using the technology for financial planning and analysis. Additionally, 68% are highly interested in using it for financial reporting and 63% for cost management and working capital optimization.
“We use it because it helps minimize manual errors in invoice processing and strengthens supplier relationship management,” the CFO of a firm in the finance and insurance sector with annual revenue exceeding $100 billion told PYMNTS Intelligence in the survey.
Additionally, the CFO of a real estate firm with annual revenue between $1 billion and $5 billion told us that agentic AI can generate “dynamic forecasts that update in real time, keeping financial projections aligned with current market trends.”
Conversely, the use of agentic AI for functions involving the movement of money is less common.
Only 32% of enterprise CFOs are highly interested in using agentic AI for treasury management. This relative wariness is likely due to both the complexity and high stakes of this function, which for a large company can involve thousands of bank accounts over dozens of countries, each with different currencies and different regulatory requirements. There is a lot that can go wrong, from incorrect sums allocated to over- or underpayments, and the consequences of a misstep can be severe. This rationale can also explain the relative disinterest in agentic AI for risk management and tax compliance across multiple jurisdictions, two highly complex areas with bottom-line consequences.
Agentic AI adoption hinges on enterprises’ willingness to grant the technology full access to sensitive internal data.
Enterprise CFOs are cautious about letting agentic AI have access to sensitive internal data. That means a direct view of internal directories (such as employee, customer or vendor lists) and action-level permissions (such as the ability to send messages, initiate payments or schedule meetings with people on those lists) to enable agentic AI to execute tasks autonomously.
In fact, no CFOs expressed a willingness to allow substantial or full access. Only 8.3% were willing to grant moderate access. While 45% were open to granting limited access, an even greater share (47%) would not grant any.
Those already using or piloting agentic AI are by far the most likely to grant access. Among these, 43% would give moderate access, and the remaining 57% would give limited access.
Read More
PYMNTS Intelligence is the leading provider of information on the consumer trends driving innovation in consumer finance, digital payments and financial inclusion. To stay up to date, subscribe to our newsletters and read our in-depth reports.
Methodology
“The CAIO Report: How Agentic AI Went From Zero to CFO Test Runs in 90 Days” is based on a survey of 60 CFOs working at U.S. firms that generated at least $1 billion in revenues last year. The survey was conducted from July 15, 2025, to July 25, 2025. The report examines the rise of agentic AI among enterprise CFOs. The sample includes executives with varying levels of strategic authority, enabling segmented analysis by sector, confidence level and risk posture. PYMNTS Intelligence designed the survey to assess expectations for generative and agentic AI performance across use cases, as well as how those expectations have evolved year over year.
