Agentic AI: What It Will Take to Move From Rhetoric to Reality

Most chief financial officers (CFOs) say they understand the concept of agentic artificial intelligence — systems that can plan, reason and take actions with minimal or no human input — but few are ready to implement the technology.

According to a July 2025 PYMNTS Intelligence report, nearly all CFOs polled were aware of agentic AI, yet only 15% expressed interest in deploying it within their organizations.

The gap reflects lingering skepticism among business leaders about the maturity and business value of artificial intelligence agents in their current form. While the technology holds promise for automating complex workflows and improving decision-making, many remain cautious amid concerns about implementation risks, oversight challenges and unproven ROI.

“A lot of companies are excited about what agentic AI can do, but not enough are thinking about what it takes to use it safely,” James Prolizo, chief information security officer at Sovos, told PYMNTS. “These tools are starting to make real decisions, not just automate tasks, and that changes the game.”

A key roadblock is lack of trust in agentic systems. According the PYMNTS report’s supplemental data, building trust depends on the following:

• User-friendly traceability: Able to provide user-friendly reports and visualizations that clearly explain the reasons behind an AI agent’s actions, with the ability to trace outputs back to the underlying input data and logic.
• Human-in-the-loop safeguards: Human-in-the-loop controls must be present to provide ongoing human supervision and enable intervention when critical decisions are being made.
• Built-in bias monitoring: Mechanisms must be present that identify and minimize bias in AI-generated content and analyses to ensure fairness and accuracy.

Without these elements, CFOs are unlikely to delegate decision-making authority to software agents, according to experts.

“Finance leaders need to implicitly trust their systems to be accurate and predictable,” Justin Etkin, co-founder and COO of Tropic, told PYMNTS. “Mainstream adoption will only occur when CFOs can see the concrete value and have confidence that these systems won’t go rogue or produce the unpredictable results that currently tarnish AI’s reputation in financial contexts.”

Etkin added that CFOs are looking for AI solutions that go beyond basic automation by offering paper trails, step-by-step actions with “undo” capabilities, and measurable time savings.

The hesitation among finance executives stands in contrast to a more optimistic outlook in other parts of the enterprise.

Raju Malhotra, chief product and technology officer at Certinia, pointed to strong momentum in the professional services sector. “Agentic AI is the most transformative addition. Autonomous digital workers can create a hybrid workforce that blends AI with human teams,” he said.

Malhotra cited Certinia’s own research showing that 83% of professional services firms are already deploying or planning to deploy agentic artificial intelligence in professional services automation within the next year.

Yet Malhotra acknowledged that even among early adopters, results are uneven. “Many organizations still struggle to see returns, with 29% stating current AI solutions fall short of expectations. The reasons are clear: a lack of internal skills and fragmented data,” he said.

See also: The Two Faces of AI: Gen AI’s Triumph Meets Agentic AI’s Caution

In financial settings, where decisions must often comply with regulatory mandates and also pass audits, the technical and cultural barriers are even higher. Integration is a major issue.

Agentic systems must connect with a wide array of internal platforms — from enterprise resource planning (ERP) software to forecasting models to compliance tools. According to Chaim Mazal, chief security officer at Gigamon, legacy IT infrastructure can become a bottleneck, especially when it comes to visibility and security.

“As AI workloads drive a surge in network traffic — doubling it in one in three organizations — traditional monitoring tools are increasingly overwhelmed,” Mazal told PYMNTS. “This makes it harder to detect when agentic AI is acting outside intended parameters, sharing sensitive data, or accessing unauthorized systems.”

Mazal said encrypted traffic and siloed authentication systems make it difficult to monitor agentic AI activity in real time. “Until companies integrate AI-aware telemetry and achieve deep, real-time observability into all data in motion, they won’t be equipped to securely govern agentic AI. Simply put: if you can’t see it, you can’t secure it.”

Beyond the technical obstacles, agentic AI also poses human and cultural challenges.

Eric Karofsky, founder of VectorHX, emphasized the importance of trust and usability. “Regardless of how sophisticated the technology or streamlined the automation, adoption comes down to whether real people can actually trust and use these systems without frustration,” he told PYMNTS. “When customers experience agentic AI that feels like a black box making unexplained decisions, their trust disappears faster than any efficiency gains can make up.”

Even in markets like payments, where automation has long played a role, agentic AI raises fresh concerns.

Edwin Loredo, partner at Core Innovation Capital, pointed to the risks of fraud, misuse of permissions and the need for stronger authentication and oversight mechanisms. “Fraud is very likely to increase. Agents can be manipulated, permissions can be exploited and bad actors will look for the edge cases in these automated systems,” he told PYMNTS.

In Loredo’s view, agentic AI will require domain-specific interfaces and rigorous logic tailored to the unique needs of industries or sectors such as finance, insurance and healthcare. “It’s unlikely a single general-purpose agent can serve all use cases well,” he said.

For now, CFOs are staying cautious. While the potential of agentic AI is widely acknowledged, few finance leaders are willing to deploy systems that operate beyond their control. As the technology matures, vendors and internal teams will need to meet higher standards for transparency, security and ROI to make adoption viable. Until then, agentic AI may remain mostly a buzzword in the boardroom.

Read more:

AI at the Crossroads: Agentic Ambitions Meet Operational Realities

The Agentic Trust Gap: Enterprise CFOs Push Pause on Agentic AI

Payments Execs on Agentic AI: ‘The Back Office Will Never Be the Same’

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

The post Agentic AI: What It Will Take to Move From Rhetoric to Reality appeared first on PYMNTS.com.