Agentic AI may be moving from the conceptual to the testing phase in corporate finance faster than most predicted.
According to a new survey from PYMNTS Intelligence of 60 CFOs of U.S. firms that generated at least $1 billion in revenues in 2024, U.S. enterprises had not reported a single live deployment or pilot of agentic AI just three months ago. However, by the start of the third quarter, more than one in 10 companies had begun using or testing agentic AI, with a growing number of CFOs saying they’re preparing to adopt it in 2026.
This rapid shift marks not only a sharp quarterly change in executive sentiment toward emerging technology, but it also suggests that finance leaders are beginning to test AI systems that can act on their own within defined boundaries. Agentic AI is still in its early stages, but the data shows the surveyed CFOs are interested in agentic AI at scale, a trend that may drive momentum and can reshape finance functions of all sizes in the near future.
Going from talk to action
Agentic AI is designed to go beyond the content creation capacity of generative AI. Instead of generating text or analysis under human control, agentic systems can take action on their own based on predefined inputs. For CFOs, knowing this difference is critical. Finance leaders, according to the data, must see agentic AI not only as an information tool but as a potential executor of tasks that traditionally required staff or layers of approval and lots of manpower
In May 2025, only 1.7% of large U.S. enterprises were even exploring the possibility of adopting agentic AI in the following year. At that time, none were piloting or using it. By July, 6.7% were using it in live settings and another 5% were testing it. An additional 8.3% were exploring adoption in the next 12 months. Altogether, just over a third (35%) of CFOs surveyed reported their firms were either using, piloting or considering the technology.
This level of change in only 90 days highlights how quickly momentum around agentic AI has built within corporate finance. According to researchers, CFO skepticism is falling at the same pace. In May, 85% of CFOs said they had no plans to adopt agentic AI. By July, that share dropped to 65%.
Unsurprisingly, technology companies have been the fastest to move forward, while service-based enterprises remain more cautious. Another key driver is how companies have used generative AI previously. Firms that deployed generative AI in finance functions like payments or risk analysis are much more likely to adopt agentic AI. One in four of those firms is already using or piloting, and half are considering adoption. On the flip side, companies that only use generative AI for low-value tasks like summarizing emails are the ones that are likely not moving ahead.
How enterprises are approaching adoption
Enterprises testing agentic AI are split on whether to build or buy. Among firms that are using or considering adoption, just over half are building in-house capabilities with their own engineering teams. That same share is also working with external partners that include teams like large AI vendors, fintechs and consultants.
The research shows companies are already running pilots and are leaning more heavily on in-house development versus vendors. Seventy-one percent of these firms are building internally, while 43% are working with external partners. Another 43% are procuring off-the-shelf tools.
Many CFOs are taking a blended approach, testing different models and strategies before committing long term. Corporate finance has already seen major accounting firms invest heavily in in-house AI development, while also acknowledging that some work will need to be outsourced as companies of similar size continue to roll out agentic AI tools.
In-house development offers self-customization and long-term control, but requires greater upfront investment and time. These are two things smaller, mid-sized and companies under private or diverse ownership structures don’t have in excess. Thus, external partnerships may allow companies of smaller stature to move fast and reduce initial costs, but come with a risk of being oversold or acquiring shelfware. This divide highlights the broader market tension between control and speed, particularly at mid-sized companies.
CFOs are also playing a central role in determining where agentive AI is applied. Seventy percent said they are very or extremely interested in using it for FP&A. Sixty-eight percent expressed high interest in financial reporting, while 63% highlighted cost management and working capital optimization.
Examples of use cases are emerging across industries:
- Education: Automating preparation of board-level financial reports
- Food and beverage distribution: Detecting anomalies in accounting records
- Healthcare: Reconciling accounts payable and receivable in real time
- Real estate: Creating forecasts that update automatically with market trends
- Technology: Identifying overspending relative to industry benchmarks
- Travel and transportation: Modeling risk tied to fuel price and currency volatility
Barriers to broader use
While CFOs see value in planning and reporting, they are more cautious about letting agentic AI handle high-stakes processes. Less than a third (32%) expressed a strong interest in areas like treasury management. Large enterprises often manage thousands of accounts across dozens of countries with different regulatory rules, where errors can lead to significant financial and legal consequences. The same caution is extended to risk management and tax compliance.
Perhaps the biggest obstacle is access. For agentic AI to operate at full capacity, it needs entry to sensitive internal directories such as client and vendor lists, along with action-level permissions to do things like send payments or schedule meetings. CFOs, many of whom have significant risk management responsibilities at their organizations, are reluctant to provide this level of control. None said they would allow full access. Only 8.3% were open to granting moderate access. Less than half (45%) would allow limited access, and 47% said they would grant none at all.
Among companies already using or piloting agentic AI, attitudes are more flexible. Forty-three percent of these firms are open to moderate access, and 57% are open to limited access. This shows that even among early adopters, a trust gap remains. Until CFOs are comfortable giving technology more autonomy, adoption will be partial.
The surge in Q3 adoption is significant but still represents the beginning of a larger story. As more companies see returns from generative AI and early agentic use cases, adoption rates are likely to continue climbing.
CFOs are not bystanders in this unprecedentedly complicated technology adoption process. They are actively testing how agentic AI can enhance planning, reporting and operational efficiency. At the same time, they must carefully weigh the risks of giving these systems broader control. How CFOs across industries navigate that balance between opportunity and caution will likely determine whether agentic AI moves from a pilot project to real-world, secure, ROI-producing technology.
