Something is happening inside financial institutions – quietly, quickly, and with very little recognition of its significance. The systems that are currently in production are not simply rule-based automation from a decade ago. They plan. They fetch data. They take sequences of actions over multiple platforms. They do not wait to be told what to do next. This is agentic AI. And the financial industry is its most consequential proving ground.
From Prediction to Action
In finance, for a long time, AI meant one thing: a prediction. Would this loan default? Was this transaction fraudulent? Each model was a prediction engine. Each model was a prediction tool. Each model stopped there.
An agentic system is different. An agentic system seeks outcomes. It does not seek outputs.
What does that mean in practice?
• What does that mean for a trading agent that watches the market, measures liquidity, and optimizes timing?
• What does that mean for a compliance agent that investigates counterparties, cross-checks against sanctions lists, and creates a preliminary report?
• What does that mean for a customer advisory agent that reviews a portfolio, detects drift, and recommends action?
The difference between prediction and action is not a minor detail. It is a fundamental shift in the role that humans play in financial decision-making.
Where the Momentum Is Building?
While adoption is not uniform, some areas have accelerated ahead, driven by the availability of data, cost considerations, and latitude for action.
Fraud prevention was first. In a world where a modern fraud agent does not simply determine if a transaction is suspicious, but instead investigates, they can, in milliseconds, analyze:
• Device fingerprint and history of locations
• Transaction velocity and merchant reputation
• Behavioral characteristics of recent transactions
This results in fewer false positives, faster decision-making, and a dramatically lower cost per case under review.
Portfolio management is the second space where agents are being adopted. The largest asset managers are now employing agents that continuously, without the need for a morning meeting, monitor positions, stress test, and execute trades in response. Human professionals are freed to focus on goal-setting, rather than moment-to-moment decision-making.
The most dramatic productivity benefits are seen in the realm of compliance. Firms using agentic AI in financial services claim to compress the entire process, including transaction monitoring and suspicious activity reporting, from days to hours, along with higher accuracy and a complete audit trail as a bonus.
Rethinking Credit from the Ground Up
Credit is where agentic AI is currently challenging assumptions that have defined the industry for decades.
Conventional underwriting is a slow, costly, and discriminatory process by design. It relies on credit bureau scores, income verification, and rule-based underwriting, a system that systematically disadvantages those with unconventional financial profiles.
The rise of agentic AI in lending is changing that narrative. An agent evaluating a small business loan today considers:
• Transaction-level cash flows and seasonality patterns
• Supplier payment history and days payable outstanding
• Industry benchmarking and macro-economic factors
• Alternative data, such as utility payments and supplier history
The result is a credit assessment that is more detailed, more current, and more forward-looking than a traditional score, allowing lenders to say yes where they previously had to say no.
Once a loan is made, the impact of agentic AI in credit is just as profound. An agent continuously monitors a borrower’s activity. If a change in revenue occurs, it will be detected weeks in advance of a missed payment, allowing a lender and a borrower to renegotiate a loan proactively, rather than reactively in default.
The Governance Gap
None of this, of course, is without its complications, and the financial industry would be foolish to pretend otherwise.
Three issues are at the heart of any serious discussion in boardrooms and risk meetings:
Accountability: Who is ultimately accountable when an agent passes on a loan request or makes a trade on its own initiative? The regulators are asking, and the answers provided by the industry have yet to be satisfactory.
Explainability: Agentic decision-making can be opaque, and for an industry that relies on a paper trail of rationale and fiduciary responsibility, a decision-making process that cannot be explained is not just inconvenient; it is a liability.
Failure at scale: While a bad decision by a human analyst may impact a few hundred cases, a bad decision by an agent may impact thousands, and a bias may be embedded in tens of thousands before anyone even realizes what is happening.
The organizations that manage this process best have a common approach.
- Maintain human decision-making above risk thresholds
- Invest in interpretability tooling alongside model deployment
- Regularly perform red teaming to identify unintended agent behaviors
- Engage regulators ahead of formal requirements
They do not see governance as a blocker to deployment. Instead, they see it as a prerequisite.
What the Next Chapter Looks Like
The pace is not slowing. The efficiency gap between early adopters and laggards continues to grow – and in a margin-compressed business, that gap is a liability.
The debate is not over whether agentic AI will reshape finance. It is over what the human component will look like on the other side.
The rosy scenario: professionals freed from data work to focus on judgment, relationships, and strategy.
The less rosy scenario: the universe of work that actually requires judgment may be smaller than the industry wants to believe.
The truth will probably be somewhere in between – and will almost certainly vary by function, institution, and market.
What is not up for debate is the direction we are headed in. The organizations that will define the next ten years are those that will treat agentic AI not as a cost-reduction opportunity, but a new business operating model, and a new business operating model requires a new business governance model.
That debate needs a lot more urgency than it is currently being given.