From APIs to Intelligent Financial Infrastructure: The Next Evolution of Open Finance

Exposing more data isn't enough anymore. The next evolution of open finance requires AI-native infrastructure that can reason over data, orchestrate workflows, based on a strict governance framework.

Executive Summary

For the past decade, the financial industry has been consumed by the plumbing of open finance. We have focused on building secure APIs, defining interoperability standards, and establishing the regulatory frameworks required to share permissioned financial data. However, exposing more data or bolting isolated chatbots onto legacy systems will not define the next era of financial services. Drawing from my experience engineering enterprise-scale systems at American Express and researching applied AI in regulated environments, it is clear that the true evolution of open finance lies in what we do with the data. We are moving from an era of data transmission to an era of autonomous reasoning. The future belongs to intelligent financial infrastructure -- AI-native systems that can reason over complex financial data, orchestrate multi-step agentic workflows, support critical decisions, and act safely within strict boundaries of governance, compliance, and human approval.

The Agentic Shift: From Data Transmission to Autonomous Reasoning

Open Banking 1.0 was characterized by static data retrieval. An API securely transmitted a customer’s transaction history from a bank to a third-party application. While valuable, this process still relied heavily on human intervention or rigid, deterministic rules engines to analyze that data and make decisions. The API was the conduit; the intelligence remained elsewhere.

Today, the landscape is shifting dramatically toward Agentic AI. Unlike traditional analytical AI or basic generative AI—which primarily assist human workers by summarizing text or flagging anomalies—Agentic AI involves single or multiple agents that can carry out tasks and make decisions autonomously [1]. These systems do not merely respond to queries; they plan, reason, and adapt in real time to handle complex, multi-step workflows.

During my time working on large-scale infrastructure at American Express, I witnessed firsthand how manual, linear operating models can become bottlenecks. This is especially evident in the fight against financial crime. Despite banks increasing their anti-money laundering (AML) and know-your-customer (KYC) spending by up to 10% annually in recent years, the industry still detects only approximately 2% of illicit global financial flows [2].

Leading financial institutions are now deploying “squads” of AI agents to orchestrate end-to-end KYC workflows. Instead of a human analyst manually gathering documents, a Data Pipeline Agent extracts and resolves entity data across multiple sources. A Research Agent analyzes corporate structures to identify ultimate beneficial owners. A Critic Agent performs quality assurance on the output before compiling a consolidated memo for human review [2]. In these agentic factories, humans are elevated from data gatherers to exception handlers and strategic reviewers.

The “Governance Illusion” and the Limits of Human-in-the-Loop

As we grant AI systems more autonomy, our approach to governance must fundamentally change. For years, the financial sector has relied on a comforting concept to manage AI risk: the “human-in-the-loop.” The assumption is that if an AI system is uncertain or produces a recommendation, a human will review and approve it, thereby controlling the risk.

However, for autonomous systems, relying solely on human-in-the-loop creates a dangerous blind spot that can be described as the “Governance Illusion.”

If the AI system itself decides what is risky, what should be escalated, and what can be executed silently, governance becomes dependent on the very system it is supposed to supervise [3]. This creates a dependency trap: if the AI’s reasoning is flawed but the system is highly confident in that flawed reasoning, the human reviewer is never alerted. The approval is logged, the workflow appears compliant, and the audit trail exists—but meaningful oversight never occurred.

Furthermore, many AI failures in financial services do not originate in the reasoning layer; they originate in the representation layer. If a customer’s identity record is fragmented, a transaction context is stale, or a consent status is ambiguous, even the most sophisticated AI will produce a confident but institutionally incorrect decision. Reviewing the final output of a flawed representation is not governance; it is liability transfer.

Building Intelligent Financial Infrastructure

To move beyond the governance illusion, financial institutions must build AI-native infrastructure designed for observability, accountability, and boundary-driven execution. The winners in open finance will invest in three critical architectural pillars:

1. Event-Driven Architecture (EDA) for Agentic Scale

Legacy financial software relies on batch processing and polling, which is fundamentally insufficient for real-time AI agents. To scale effectively, AI agents require Event-Driven Architecture (EDA). EDA allows agents to subscribe to state changes -- such as a flagged transaction or a modified consent preference -- and react instantly without waiting for a scheduled job [4]. Crucially, if an agent fails, event persistence ensures that no data is lost, allowing the agent to resume precisely where it left off.

2. AI Observability and Boundary-Governed Execution
We must transition from human-in-the-loop to boundary-governed AI. Humans should not mechanically approve hundreds of AI recommendations; instead, they must define the strict conditions under which AI is permitted to act—where autonomy is allowed, which actions must be reversible, and which decisions require independent verification [3]. This requires robust AI observability. Observability tools must detect model drift and ensure that AI systems operate within the guardrails defined by frameworks like the U.S. Treasury’s Financial Services AI Risk Management Framework (FS AI RMF) [5].

3. Open Standards and Common Controls
The complexity of AI risk cannot be managed in isolation. The industry requires standardized, technology-neutral controls that enable consistent governance across institutions and jurisdictions. Initiatives like the Fintech Open Source Foundation’s (FINOS) “Common Controls for AI Services,” launched in collaboration with global banks and cloud providers, represent exactly this kind of collaborative foundation [6].

Conclusion

The first decade of open finance was about unlocking data. The next decade will be about orchestrating it intelligently, safely, and at scale. As embedded finance and smart data initiatives expand, competitive advantage will no longer stem simply from API access. It will stem from execution speed, precision, and trust. Financial institutions that treat AI as a bolt-on tool or rely on superficial human oversight will struggle to manage the complexity of modern markets. The true leaders will be those who recognize that AI is not just an application; it is the new infrastructure. Building intelligent, observable, and strictly governed financial systems is not a technology investment alone -- it is a strategic commitment to the kind of trust that financial services ultimately depend upon.

References

[1] Sydorenko, Igor. “Agentic AI in Financial Services: A Research Roundup for 2026.” Neurons Lab, January 30, 2026. Online [Available] https://neurons-lab.com/article/agentic-ai-in-financial-services-2026/

[2] Verhagen, Alexander, et al. “How agentic AI can change the way banks fight financial crime.” McKinsey & Company, August 7, 2025. Online [Available] https://www.mckinsey.com/capabilities/risk-and-resilience/our-insights/how-agentic-ai-can-change-the-way-banks-fight-financial-crime

[3] Singh, Raktim. “The Governance Illusion in Financial Services: Why Human Oversight Is Not Enough for Autonomous AI.” Finextra, May 19, 2026. Online [Available] https://www.finextra.com/blogposting/31758/the-governance-illusion-in-financial-services-why-human-oversight-is-not-enough-for-autonomous-ai

[4] Atlan. “Event-Driven Architecture for AI Agents: Patterns and Benefits.” March 16, 2026. Online [Available] https://atlan.com/know/event-driven-architecture-for-ai-agents/

[5] U.S. Department of the Treasury. “Treasury Releases Two New Resources to Guide AI Use in the Financial Sector.” February 19, 2026. Online [Available] https://home.treasury.gov/news/press-releases/sb0401

[6] FINOS. “Global Financial Institutions and Technology Leaders Collaborate Under FINOS to Launch Open Source Common Controls for AI Services.” June 24, 2025. Online [Available] https://www.finos.org/press/global-financial-institutions-and-technology-leaders-collaborate-under-finos-to-launch-open-source-common-controls-for-ai-services