AI Agents in Financial Services: From Pilot to Production

AI pilots are everywhere. AI agents you'd trust with real payments are not.

Most systems store data as a snapshot. They know the current state, but not how or why it got there. That worked when systems only displayed and updated records. AI agents raise the stakes. To make agents trustworthy, explainable, and production-ready, you need the full story: what happened, when, and why.

The good news is that much of that history already exists in the systems you run today, waiting to be surfaced as business events—no rewrite required.

In this fireside chat, Fu Sing Voon and Timothy Lam, backend engineers from MoneyLion, a Gen Digital company, who run AI agents against real payment operations, join Allard Buijze, CTO of Axoniq, to talk about what separates a working demo from a production system. No slideware, just practitioners comparing notes.

What we'll cover

• The demo-to-production gap. Why it's an engineering problem (testing, guardrails, edge cases), not a model problem.

• Facts vs. guesses. What an agent needs to explain why a payment is stuck, and how to stop it from confusing what it knows with what it assumes.

• When the agent gets it wrong. How to trace what it did, contain the damage, and reconcile.

• Working with existing systems. Why agents get limited tools and APIs, and never direct database access.

• Earning autonomy. The path from read-only assistance to human-approved actions to limited autonomy, and what earns an agent the next step.

Who should attend

CIOs, Heads of IT, and technology leaders in banks, payments, and financial services, plus the architects, engineering leads, and data teams working close to core systems