The global financial system is undergoing a seismic transformation. For decades, the industry has been defined by human traders on bustling floors, relationship managers poring over client portfolios, and armies of analysts meticulously reviewing credit applications. While technology has incrementally automated certain tasks, the fundamental assumption remained: a person would log in, navigate a screen, interpret data, and decide what happens next . That assumption is now being challenged. Agentic AI is stepping in, transitioning from a passive tool that generates answers to an active, autonomous actor that plans, executes, and adapts to achieve defined goals . According to the Cambridge Centre for Alternative Finance’s 2026 Global AI in Financial Services report, the industry is rapidly outpacing regulators in AI adoption, with agentic AI emerging as a highly accessible frontier . This is not just an incremental upgrade; it is the dawn of a new era where the primary users of core financial infrastructure may no longer be human.
The Great Refactor: From Human-Centric to Agent-Ready Systems
The move to agentic AI represents a fundamental architectural shift for financial institutions. Current banking systems are predominantly designed around human users and their workflows, relying on graphical user interfaces and manual interpretation. The new design question, however, is not about access to better AI models but about whether an institution’s data and IT systems can be discovered, understood, operated, and audited by AI agents . Systems that cannot be used by AI agents risk becoming functionally obsolete, leading to what experts call a “Great Refactor” the race to turn legacy software into platforms that AI agents can actually use .
Large financial institutions are particularly challenged by legacy technology estates built on decades-old code, undocumented rules, and fragmented data stores. AWS has found that in many cases, over 40% of business rules are embedded solely in code with no supporting documentation, creating a new form of “agent-blind technical debt” . This debt makes it impossible for AI agents to interpret and operate systems effectively, as they cannot rely on the institutional memory that human workers often possess. The construction of a new financial operating system is necessary one that transitions from deterministic, hand-coded logic to model-driven, natural-language-mediated, and agentic architectures .
A New Set of Design Principles
To address these hidden constraints, industry leaders are developing a new set of design principles for agent-ready systems. These can be categorized as follows:
A. Secure Access Pathways: Agents must be given access to systems through secure and controlled pathways, ensuring that their operations do not compromise sensitive data or system integrity.
B. Machine-Readable Annotations: Systems must be annotated so that data, rules, and constraints are machine-readable. This allows AI agents to understand the context and limitations of the data they are working with.
C. Actionable Capabilities: Capabilities need to be actionized, meaning agents can call tools and APIs to execute tasks directly, rather than simply reading information. This moves agents from passive observers to active participants.
D. Intent-Driven Workflows: Workflows should augment human teams by assembling steps around intent. Instead of navigating complex menus, users describe the outcome they want, and the system assembles the necessary steps .
E. Abstracted Business Logic: Business logic needs to be abstracted into policies and constraints that agents can reason over, allowing for more flexible and intelligent decision-making.
F. Autonomous Lifecycle Components: Where appropriate, parts of the lifecycle can be autonomized with clear governance, testing, and oversight, allowing agents to operate independently within safe boundaries .
Real-World Use Cases and Applications
The theoretical promise of agentic AI is rapidly translating into real-world applications across the financial services spectrum. The technology is being deployed to automate labor-intensive functions that were previously the exclusive domain of highly skilled (and expensive) human workers. According to a report from Moody’s, Research Assistant users are consuming 60% more research while cutting task completion times by 30%, with over 90% of AI interactions now focused on high-value analytics . This represents a fundamental shift in how financial professionals allocate their time and energy.
Anthropic recently launched 10 new AI agents tailored for banks, insurers, asset managers, and fintech firms through its Claude ecosystem. These agents are designed to automate critical functions, including pitchbook preparation, earnings analysis, financial statement reviews, credit memo drafting, KYC verification, accounting reconciliation, and valuation assessments . The company has also expanded integrations with major financial data providers like FactSet, Moody’s, Morningstar, PitchBook, and S&P Capital IQ, allowing the AI agents to work with structured financial datasets rather than relying solely on open internet information. This approach significantly improves reliability and reduces hallucination risks in financial analysis .
Goldman Sachs is working with Anthropic to develop AI agents to speed up trade and transaction accounting, client due diligence, and onboarding – core areas where banks depend on data, controls, judgement, and execution working together . Citi is using AI to migrate data from legacy systems and automate coding, with CTO David Griffiths saying the agentic approach can be 2 to 20 times faster for certain tasks . Morgan Stanley has used its DevGen.AI tool to review 9 million lines of legacy code and translate them into plain-English specifications, turning old logic into something that can be understood, explained, and modernized . In Asia, Bank of Singapore is already using agentic AI in KYC, with an assistant drafting Source of Wealth reports and reducing cycle times from days to hours .
Key Use Cases
The following are six key use cases demonstrating the power of agentic AI in the financial sector:
A. Detecting and Preventing Fraud in Real Time: Fraud detection agents can monitor transaction patterns in real time, learn from new types of fraud, and take immediate action by alerting teams or freezing suspicious accounts, largely without human intervention . This is a significant step up from traditional rule-based systems that often produce high false positive rates.
B. Expanding Credit Access with AI-Enabled Underwriting and Lending: Agentic AI can orchestrate the complete loan origination process by leveraging open banking to gain a complete understanding of customers’ financial profiles. This transforms underwriting from a days-long process to one that takes minutes, all while expanding access to credit and reducing default risk .
C. Managing Wealth Portfolios Autonomously: Agentic AI can deliver hyper-personalized wealth management at a scale that human advisors cannot match, significantly reducing operational costs and potentially democratizing wealth management for mainstream audiences . This opens new revenue streams for firms looking to move beyond ultra-high-net-worth clients.
D. Always-On Compliance and Regulation Monitoring: Keeping up with changing regulations costs tier-one banks in excess of $1 billion a year . Regulatory AI agents can monitor and implement compliance in real time, working alongside human subject matter experts to introduce effective controls and raise alerts before they become critical.
E. Developing and Delivering New Digital Products Faster: Financial firms are using an AI-powered software development life cycle (SDLC) to take on the more tedious, time-intensive parts of software development. This allows developers and engineers to focus on architecture and oversight rather than routine coding .
F. Credit Risk Modeling and Validation: As demonstrated in recent academic research, agentic crews comprising a manager and multiple specialized agents can effectively perform complex tasks such as exploratory data analysis, feature engineering, model selection, hyperparameter tuning, model training, and even model risk management (MRM) . This approach promises to revolutionize how financial institutions build and validate their core risk models.
The Governance Imperative: Taming the Autonomous Beast
While the opportunities are vast, the autonomous nature of agentic AI introduces significant risks that demand robust governance frameworks. Unlike static automation tools, agentic systems act on their own, planning and executing multi-step tasks without continuous human approval. This creates the potential for unauthorized or illegal actions, data breaches, cascading errors, and disruption to connected systems . The Financial Stability Board (FSB), a global standard-setter, has strongly encouraged financial firms to implement safeguards to mitigate these risks. The FSB has outlined a series of proposed “sound practices,” urging firms to define clear boundaries on AI use, embed safeguards, and even consider adapting HR controls and processes to treat AI agents as “synthetic employees” .
Financial institutions are responding by prioritizing governance over speed in agentic AI deployments. According to a 2025 industry study, 99% of companies plan to move AI agents into production, yet only 11% have successfully done so . This gap is largely due to the difficulty of governing these systems in a regulated environment. Industry experts increasingly argue that “human-in-the-loop” (HITL) frameworks are becoming the core architecture standard for production AI systems. Rather than acting as a compliance formality, human oversight is now being embedded directly into system design. Research from Gartner’s 2025 AI Governance Survey found that enterprises using structured human oversight protocols experienced 47% fewer AI-related incidents and achieved significantly faster internal adoption compared to firms deploying fully autonomous systems .
To address these challenges, researchers have proposed comprehensive governance frameworks like TRACE (Trust, Review, Accountability, Critique, Explainability). TRACE operationalizes governance through a layered structure encompassing Governance & Compliance, Operational Agents, and Oversight & Assurance, embedding measurable indicators and policy-aligned controls at each layer . This approach provides a replicable blueprint for financial institutions seeking to deploy governance-compliant, trustworthy, and auditable multi-agent AI automation.
Essential Governance Principles
To implement responsible agentic AI, financial institutions should commit to the following principles:
A. Explainability and Transparency: Agentic AI should not operate as a black box. The AI agent should be able to explain its work and recommendations, providing complete transparency into what they are based on. An automated underwriting and lending workflow, for example, should be able to explain and trace application review outcomes so humans can ensure they are making responsible credit decisions .
B. Strong Data Security and Privacy: Companies that are using publicly available AI platforms should take extra precautions to eliminate any risk that sensitive, proprietary, and private information could be leaked. This may require building and/or training their own models and solutions in a private cloud to ensure strict controls and oversight .
C. Human-in-the-Loop Oversight for High-Risk Actions: Firms should not use AI output without human validation for irreversible actions involving payments, account decisions, infrastructure changes, or customer risk assessments. A hybrid expert-in-the-loop approach, where AI makes recommendations and people decide which actions to take, enhances accuracy, efficiency, and interpretability .
D. Robust Evaluation Frameworks: Financial institutions must develop hybrid evaluation strategies that blend human review, automated testing, and LLM-as-judge approaches to balance accuracy, cost, and speed. Bloomberg’s ASKB agent, for example, primarily uses human expert-driven evaluation to ensure high quality and accuracy in its financial analysis .
E. Clear Orchestration and Role Definitions: For multi-agent systems, it is crucial to define clear orchestration and role definitions for subagents. Each agent should have a named accountable owner to avoid blurred responsibility .
The Competitive Divide: Fintechs vs. Incumbents
The race to agentic AI is creating a new competitive divide in the financial services industry. The Cambridge Centre for Alternative Finance’s research found fintechs ahead of traditional financial institutions in agentic AI adoption, at 57% versus 45% . Fintechs and newer firms may have cleaner stacks, fewer legacy constraints, and more flexibility to design around agentic workflows from the start. They are the nimble startups that can build their systems from the ground up with AI agents in mind, giving them a significant speed advantage. However, incumbent banks possess their own strengths: scale, vast amounts of proprietary data, established trust, and deep client relationships . While they have more complex estates to refactor, the ability to leverage their rich data sets could provide a powerful competitive moat.
The next banking transformation is not just about adopting AI. It is about preparing for a market in which more financial activity is initiated, interpreted, and coordinated by software. The most competitive institutions will not simply be those with the best apps or the largest AI teams. They will be those whose systems can be safely discovered, understood, operated, and audited by agents. The next users of banking infrastructure may not be human. Banks that prepare for that shift will be better placed to move faster, serve clients more intelligently, and preserve their relevance in the next generation of digital finance .
Conclusion: The Future is Autonomous, Accountable, and Agentic
Agentic AI is not a distant concept; it is an operational reality rapidly reshaping global finance. The 2026 Global AI in Financial Services report anticipates that 81% of firms expect autonomous agents to be meaningfully achieved by 2030 . The transition promises unprecedented efficiency gains, from automating the entire loan origination process to autonomously managing wealth portfolios and detecting fraud in real-time. However, this transformation is contingent upon a fundamental “Great Refactor” of legacy systems and the establishment of robust governance frameworks that ensure accountability, transparency, and human oversight. The future of finance is undeniably agentic, and the institutions that can successfully navigate this complex landscape will be the ones that thrive in the decades to come. The journey ahead requires a delicate balance leveraging the immense power of AI agents while maintaining the human judgment and regulatory compliance that are the bedrock of a stable financial system.
