October 1, 2026
The financial services sector stands at a critical technological crossroads. While Generative AI (GenAI) initially captured the imagination of boardrooms with its ability to draft emails and summarize reports, a new, more ambitious paradigm has emerged: Agentic AI. Unlike passive models that merely respond to prompts, agentic systems are designed to perform complex, multi-step tasks, make autonomous decisions, and interact with enterprise software ecosystems to drive tangible business outcomes.
However, as the initial fervor surrounding this technology begins to settle, a stark reality has emerged: the gap between experimental pilot programs and full-scale enterprise production remains wide. To bridge this divide, industry leaders are shifting their focus from raw model power to the foundational necessity of "enterprise context."
The State of the Industry: A New Roundtable Discussion
On October 1, 2026, GlobalData hosted a high-level expert roundtable titled "Agentic AI in Banking & Financial Services: Beyond the Hype." The session convened industry veterans to dissect the friction points hindering widespread adoption. Panelists included Jeff Veis, Chief Marketing Officer at Impetus Technologies; Deepak Khosla, Chief Growth Officer & Head of AI Business at Impetus Technologies; and Stephen Walker, a leading Retail Banking Analyst at GlobalData.
The discussion sought to demystify why, despite significant investment, many financial institutions find themselves stuck in a state of "perpetual experimentation."
Chronology of the Agentic Shift
The trajectory of AI in banking has been rapid, moving through distinct phases that have led to the current focus on agency:
- 2023: The GenAI Explosion: Financial institutions rushed to implement basic Large Language Models (LLMs) for employee productivity, such as document summarization and drafting customer service responses.
- 2024: The Era of RAG (Retrieval-Augmented Generation): Banks realized that off-the-shelf models were prone to hallucinations. The focus shifted to RAG to provide models with better access to internal documents.
- 2025: The Rise of Agentic Frameworks: Industry leaders began experimenting with autonomous agents capable of chaining tasks—moving beyond "chatting" to "doing."
- 2026: The "Production Gap" Realization: As seen in the GlobalData roundtable, the current year is defined by a sobering realization: AI agents are only as good as the enterprise data and governance policies that frame them.
The High Stakes of Autonomous Finance
During the roundtable, Deepak Khosla provided a critical distinction that defines the current challenge. "An agent in banking is not just summarizing a document," Khosla noted. "It could influence and impact credit, fraud, payments, customer treatment, reporting, or advice. The bar for production is therefore much higher in the banking and financial services sector."
This shift in scope changes the risk profile entirely. In other industries, a "hallucination" by an AI agent might result in a slightly inaccurate marketing email. In banking, an agent that misinterprets a regulatory compliance rule or makes an unauthorized payment could result in systemic risk, heavy regulatory fines, and permanent reputational damage.
Key Barriers to Production Adoption
The roundtable participants identified five primary obstacles preventing financial institutions from moving agentic AI out of the sandbox:
1. Lack of Data Readiness
AI agents require more than just access to data; they require "AI-ready" data. This means data that is cleaned, tagged, and accessible in real-time. Much of the data in legacy banking systems remains siloed or trapped in monolithic architectures that are not conducive to the fluid reasoning required by agentic systems.
2. Fragmented Enterprise Context
An agent cannot function effectively if it lacks a holistic view of the customer or the business process. If an agent is tasked with a mortgage approval workflow but cannot cross-reference historical credit interactions, real-time risk scoring, and internal governance policies, its decision-making will be inherently flawed.
3. Governance and Safety Gaps
Traditional AI governance focuses on model performance and fairness. Agentic AI, however, requires "behavioral governance." Institutions must define the guardrails for what an agent is allowed to do, how it should escalate a decision to a human, and how it should log its reasoning for audit trails.
4. Regulatory Expectations
Regulators are increasingly scrutinizing AI deployment in finance. Banks are being asked to demonstrate that they can explain the "why" behind an AI-driven decision—a feat that is notoriously difficult with complex, multi-agent systems.
5. Technical Integration
Many institutions struggle to integrate AI agents into their existing operational stack. Without deep integration into the core banking platforms, agents remain "islands of intelligence" that cannot execute the final steps of a transaction.
The Strategic Pivot: Engineering Enterprise Context
The consensus among the experts at the GlobalData event was clear: the solution to these barriers is "context engineering."
According to the panel, successful agentic AI adoption depends on grounding these systems in the granular realities of the business. This involves moving away from simply feeding raw data to models and toward building robust semantic layers, ontologies, and knowledge graphs.
"Our approach starts with the belief that agentic AI success depends on the quality of enterprise context available to agents," Khosla explained. "If that context is fragmented, stale, or poorly governed, they will fail in production."
Building a Foundation for Trust
To achieve this, banks are advised to invest in:
- Semantic Layers: Standardizing data definitions so that an "account balance" means the same thing to the AI as it does to the general ledger.
- Knowledge Graphs: Mapping the relationships between customers, products, and policies to allow agents to "reason" across disparate data sets.
- Trusted Data Pipelines: Ensuring that the information the agent uses is verified, up-to-date, and compliant with data privacy regulations.
Implications for the Future of Banking
The transition to agentic AI is not merely a technical upgrade; it is a fundamental shift in how banks operate.
Measurable ROI
By focusing on "context-first" deployment, banks can move toward high-impact use cases where the Return on Investment (ROI) is quantifiable. This includes automated loan processing, proactive fraud detection that understands individual customer patterns, and hyper-personalized financial advice that accounts for the full scope of a client’s portfolio.
The Human-in-the-Loop Paradigm
The future of banking is likely to be a "hybrid intelligence" model. Agents will handle the heavy lifting of data synthesis and routine decision-making, while human advisors and operators act as the final authority, overseeing the agents and handling high-complexity, high-emotion scenarios.
Regulatory Compliance as a Competitive Advantage
Institutions that successfully build robust governance and explainability into their agentic workflows will likely gain a competitive advantage. As regulators tighten requirements for AI, those with transparent, audit-ready agentic systems will be able to scale their operations faster than competitors who are forced to pause for compliance audits.
Conclusion: Moving Forward
The excitement surrounding Agentic AI is justified, but it is currently outpacing the infrastructure required to support it. As the GlobalData roundtable concluded, the next phase of the AI revolution in banking will be won not by those with the largest models, but by those with the best-engineered enterprise context.
For financial institutions, the path forward is clear: move beyond the hype, invest in the data foundations that make intelligent agents safe and effective, and prioritize the structural integrity of the enterprise context. The banks that succeed will be those that treat AI agents not as black-box solutions, but as integrated components of a well-governed, data-driven financial ecosystem.
For those seeking to navigate these challenges, the full GlobalData roundtable provides a deeper look at the technical frameworks required for safe and effective deployment. Industry professionals are encouraged to review the session to better understand how context engineering can serve as the bedrock for their organization’s AI strategy.
