The clearest example in financial services today is fraud management. An anomaly is detected, action is initiated and the customer experience becomes a quick confirmation rather than an embarrassing decline or a lengthy dispute process. That’s not automating a step. It’s removing the step entirely.
As i2c CEO Amir Wain argues in a new PYMNTS eBook, "Building the Agent-Ready Payments Enterprise", this shift marks the third era of AI: the move to agentic AI. Most companies, however, are asking the wrong question.
They focus on which model to deploy. The real obstacle is infrastructure. "The harder question," Wain writes, "is whether the infrastructure underneath was ever designed to let something act on its own."
Why Most Corporate AI Investments Are Already Flawed
The market is flooded with pitches for advanced large language models promising to revolutionize workflows. Teams rush to deploy chat interfaces and copilots, expecting a productivity boom. The results are often underwhelming. Wain, whose company provides a global payments processing platform, pinpoints the disconnect.
"When AI is layered onto fragmented systems and disconnected data, what looks like an AI initiative quickly becomes a data integration project."
In other words, agents need clarity to act. An AI agent can only orchestrate across systems it can see and only act in real time if the underlying platform responds to events as they happen. A disjointed technology stack, built over decades with siloed data models and manual handoffs, is incapable of supporting true autonomy. The first step isn't picking a model, it's examining the foundation. This foundational challenge is why the winners in the transaction data race are poised to dominate, as we've explored in our analysis of B2B payments AI winners.
From Suggesting to Doing: How Agentic AI Actually Works
To understand the leap, it helps to trace the evolution. Wain identifies three AI eras: rule-based systems in the 1980s, machine learning for pattern detection starting around 2010, and now agentic AI. Generative AI delivered reasoning. "Agentic AI adds perception, action and memory," he writes.
This means an agent is not a chatbot. It is a persistent, goal-oriented system that:
- Perceives events and data in real time from connected systems.
- Reasons about the appropriate next action based on policy and context.
- Acts autonomously, executing a task within a digital environment.
- Learns from every outcome to refine future decisions.
The operational impact is profound. Decisions that required analyst review and a call-center process can be handled within the transaction flow itself. The workflow doesn't get faster, it gets shorter.
The Unseen Battleground: Architecting for Autonomy
The competitive edge in this new era won't come from having the most sophisticated AI strategy. It will come from architectural readiness. "The companies winning in this era aren't necessarily the ones with the most sophisticated artificial intelligence strategy," Wain notes. "They're the ones whose architecture was built to support it."
For i2c, decisions made years ago to build a unified data model are now paying off. Their fraud systems can read signals across a customer's entire relationship, not just a single transaction. This allows them to "move decisions closer to the transaction itself."
This is the critical, unglamorous work. It involves:
- Unified Data Models: Ensuring clean, structured, and accessible data across all relevant systems.
- Event-Driven Architecture: Building platforms that respond to events as they happen, not in batch cycles.
- API Standardization: Creating machine-readable interfaces that allow agents to perceive and act.
Without this foundation, an AI agent is like a brilliant strategist with no army, no map, and no communications gear. Its potential is theoretical. This architectural shift is forcing a fundamental redesign of core payment logic across the industry.
Governance Is the New Control Tower
Autonomy introduces a new kind of risk. Systems are now making judgment calls that were once exclusively human.
"As autonomy increases, decision rights shift. Judgments that once belonged to people increasingly belong to systems, with humans positioned to intervene rather than approve every action."
Getting this balance is critical. Too much human oversight negates the speed benefit. Too little creates blind spots where errors can cascade. This is where governance moves from an IT policy to a core operational function.
Wain is explicit: "Permissioning, audit trails and human oversight cannot be afterthoughts. The organizations scaling agentic AI successfully are building those guardrails before they need them, not after an incident exposes the gaps."
These guardrails must be coded into the system's architecture. They include spending limits, ethical boundaries, escalation protocols, and clear audit logs that provide a "black box" for every agent decision. This isn't a bolt-on feature, it's a design principle.
Build, Buy, or Partner? A Reset for Enterprise Strategy
The arrival of mature agentic platforms is also resetting the classic build-versus-buy calculus. Wain suggests a pragmatic reset.
"Unless an organization operates at the scale of the largest players in its industry, building proprietary AI infrastructure from scratch is rarely the best use of capital or talent. The better question is which capabilities truly differentiate your business and which have already been solved well by trusted partners."
The new strategic question becomes: Where does autonomous action provide us a unique advantage, and where is it a commodity? You build the agent that manages your proprietary supply chain logic. You partner with a specialist for anti-money laundering monitoring. This "hybrid" approach mirrors a broader trend where card networks are stretching beyond payments to provide these value-added services.
What To Do Next: Audit Your Tech Stack for Agent Readiness
For enterprises watching this shift, the imperative is clear. Your first move is not an RFP for an AI vendor. It is an internal audit.
- Map a Single Painful Workflow. Choose a universal irritant like vendor onboarding, travel expense reconciliation, or customer dispute handling.
- Identify Every Handoff and Decision Point. Chart where humans move data between systems or make a yes/no judgment. Each of these points is a fracture an agent must bridge.
- Evaluate Data Accessibility. Can a single software entity, with the right permissions, access all the data needed to complete this workflow? Is that data consistent and clean?
The goal is to find where your systems don't talk to each other. Because that's where your real project begins. If agents are the pilots, your job is to ensure the control tower, the runway, and the navigation systems are agent-ready. Partnerships with companies like Plaid are starting to solve this by letting AI agents tap bank data mid conversation, bridging a critical data gap.
The final bar for evaluating any AI use case hasn't changed, it's just higher: economic benefit, structured data, real scale, and auditability. As Wain concludes, "Agentic AI doesn't lower that bar. It raises the stakes, because these systems aren't just advising anymore. They're acting." The organizations that prepare their foundations now will be the ones for whom work doesn't just get faster, it quietly disappears.
Why This Changes Everything
- AI agents can eliminate entire operational workflows from fraud management to disputes, turning complex processes into seamless, immediate user confirmations.
- Most corporate AI investments fail because they focus on model selection over infrastructure, revealing that a cohesive, real-time data platform is the true prerequisite for agent success.
- In financial services, the companies that win will be those that build their systems around real-time transaction data and agentic orchestration, not those that merely add AI layers to legacy technology.
Originally published on XOOMAR. For more news and analysis, visit XOOMAR.
Top comments (0)