Phani Kumar Pendurthi

Roadblock Title:

Building an Agentic Payment Flow: Architecture Patterns and Pitfalls

Time:

Tuesday - 1:00 PM (Tower A)

Abstract:

AI agents are now authorized to browse, decide, and buy — but most payment infrastructures were never designed for them. I’ll walk through the real architectural challenges of building payment flows that agents can reliably invoke: designing MCP-compatible APIs, handling idempotency when the “user” is a non-deterministic model, and enforcing authorization boundaries when an agent acts on behalf of a human.

Drawing from hands-on experience in enterprise-scale payment systems, I’ll share what breaks first, what patterns hold up in production, and what every fintech engineer must consider before exposing payment capabilities to an agent runtime.

No theory — just patterns, pitfalls, and guidance you can take back to your codebase.

Bio:

I’m a Principal Software Engineer at Mastercard with 18 years of experience in software engineering, data analytics, and distributed large-scale systems. I’ve built and optimized fintech and payment architectures globally across payments, telecom, banking, and manufacturing, working across multiple countries and mentoring engineers along the way. I’m now exploring the frontier of agentic commerce — designing systems where AI agents can autonomously initiate, authorize, and complete financial transactions at scale, at the intersection of enterprise architecture, MCP, and AI-driven payment flows. I’m passionate about building systems that are scalable, high-performance, and reliable — and helping engineers navigate the shift from human-driven to agent-driven financial systems.