Weighing Agentic AI Promise with Existing Reality
- Roy Urrico
- 33 minutes ago
- 4 min read
By Roy Urrico
Many financial services firms have been quick to deploy agentic AI, which promises to accelerate development cycles, automate complex work, and make IT teams more productive. Without a clear strategy, however, organizations risk creating AI sprawl, deploying agents whose decision processes are difficult to explain, and introducing inconsistent experiences for employees and accountholders. That’s the consensus of 1,900 global IT leaders surveyed by OutSystems for its 2026 State of AI Development report.
The research, collected December 2025 through January 2026, also found that 71% of credit unions and banks are already using agentic AI to build data and analytics applications. However, there is a clear gap between adoption and what actually works in practice.
Boston-based OutSystems provides an agentic systems platform built for the enterprise. “We help organizations rapidly build mission-critical applications and agentic systems, modernize legacy processes, and govern AI at scale on a single platform,” said Gonçalo Borrêga, Vice President of Product, AI and AppDev at OutSystems, who sat down with Finopotamus to discuss the results of the report and AI implementation strategies that are proving successful.
“In highly regulated industries like financial services, where organizations cannot sacrifice security or compliance for speed, our unified, governed approach to AI and software development is essential,” Borrêga proposed. “At OutSystems, we’re not just helping organizations experiment with AI: we’re empowering them to turn AI into production-ready systems that connect data, workflows, APIs, and user experiences in a governed way. For enterprises, that combination of speed and control is what makes AI useful in practice, not just interesting in theory.”
Facing Barriers to AI Success
“Financial institutions are operating under some of the toughest constraints in any industry,” explained Borrêga. “They are managing sensitive member and customer data, strict regulatory obligations, legacy core systems, and high-stakes decisions where accuracy, auditability, and human oversight still matter.”
The challenge is not a lack of interest in AI – it is that success requires more than deploying a model, noted Borrêga. “Institutions need AI to work across fragmented data environments, integrate with core systems, comply with evolving regulations, and fit into existing business processes without increasing risk, which is why governance is paramount.”
The State of Artificial Intelligence Development report also found that 94% of organizations are concerned that AI sprawl is increasing complexity, technical debt, and security risk, while only 12% are using a centralized platform to manage AI. “In financial services, those gaps are amplified because the risks associated with uncontrolled AI are exceptionally high,” said Borrêga.
Research Findings About FIs Using Agentic AI
Market momentum for agentic AI continues to build, with 97% of report respondents saying they are exploring agentic strategies right now, survey findings also revealed. “While leaders are eager to advance from pilot to production, their data and governance frameworks aren’t set up for agentic artificial intelligence. Only 36% of respondents have a centralized approach to AI governance, with most relying instead on project-level rules or ad hoc approaches,” the report found.
The research also showed financial services organizations are moving quickly from experimentation to production with agentic AI, particularly in data and analytics use cases, according to Borrêga. He said, “In fact, 71% of financial institutions are already using agentic AI to build these applications, yet only 12% of IT leaders report achieving a 76-100% success rate, highlighting a clear gap between adoption and what actually works in practice.”
Borrêga sees success in focused, high-value use cases like accelerating document processing, improving data workflows, and supporting human decision-making in areas like underwriting, servicing, and fraud detection. “The common thread is that AI delivers value when it’s embedded into governed systems that bring together data, workflows, and oversight, rather than deployed as standalone tools.”
A Gap Between Adoption and What AI Actually Works
A gap exists because adoption is easy to measure, but operational value is much harder to deliver, indicated Borrêga. “It is relatively simple for an organization to pilot an AI tool or launch an agent in one part of the business, but it is much harder to make that system reliable, secure, explainable, integrated, and scalable across the enterprise.”
Organizations are moving quickly toward agentic AI, but governance, architecture, and operating models are still catching up, Borrêga emphasized. “Only 36% have a centralized AI strategy, and just 12% are using a centralized platform to manage sprawl. So, the issue is not whether institutions are adopting AI, but whether they have the foundation to make AI repeatable and trustworthy in production.”
In practice, Borrêga said, “AI works when it is applied to a clear business problem, equipped with the right data, constrained by the right guardrails, and introduced with the right degree of human supervision. The organizations seeing the best results are not chasing AI for its own sake. They’re focusing on practical use cases that improve efficiency, reduce manual work, and enhance customer experiences.”
