Mr. Cherukuri of Navy Federal Credit Union, USA

Generative AI in the enterprise has reached an awkward middle age. The demos are impressive, the pilots are plentiful, and yet most organizations still struggle to put an AI agent anywhere near a process that actually matters — a loan decision, a claim, a dispute, a servicing request. The gap is not model capability. It is trust, traceability, and cost. This session looks at how Pega approaches that gap, and what it means for teams building on the platform today.

We will start with where generative AI already lives inside Pega: as design-time acceleration rather than runtime guesswork. Blueprint turns a described process into a build-ready application design, Infinity Studio brings AI into the development environment itself, and Knowledge Buddy grounds answers in an organization's own content instead of an open-ended model. The through-line is Pega's Predictable AI architecture, which pushes the heavy reasoning into design time so that runtime stays token-efficient and repeatable.

From there we move to agents. Infinity 26 lets developers connect and orchestrate both Pega and third-party agents inside governed workflows, so agents can join mission-critical processes with control, compliance, and predictable outcomes. Agentic Process Fabric provides the orchestration layer, with support for Model Context Protocol and Agent-to-Agent so agents can coordinate across applications and platforms, and with security controls that regulate which agents talk to whom. MCP now works in both directions: external harnesses and assistants can invoke Pega workflows, and Pega can reach outward. We will unpack what that architecture buys you — and what it asks of you in exchange.

The session then turns to the uncomfortable part. Agentic projects fail for mundane reasons: unbounded token spend, inconsistent outputs, no audit trail, and code nobody wants to own. Pega's own research put the defect rate of AI-generated code at roughly 1.7 times that of human-written code — tolerable in a prototype, unacceptable in a system that pays or denies a customer. We will look at where a deterministic workflow should stay deterministic, where an LLM genuinely adds value, and how human-in-the-loop checkpoints should be designed rather than bolted on afterward. Pricing gets the same treatment: Pega's flat fee per resolved case rather than per token changes the economics of scaling, and it is worth understanding why.

We will close with a worked example — an agent built to analyze business requirements and produce level-of-effort estimates, grounded in a real connector catalog and real estimation rules. It is a small, unglamorous use case, and that is the point: it shows how knowledge grounding, guardrails, and honest scoping matter more than model choice, and why the right framing for these agents is assistance to expert practitioners rather than replacement of them.

Attendees will leave with a clear map of Pega's GenAI and agentic capabilities, a set of criteria for deciding which work is a good candidate for an agent, and a realistic view of the governance and cost questions to answer before anything reaches production.

Biography:

Experienced Software architect with over 17 years of experience in design and develop and deliver software applications using AI , python , Java and Pega. Currently working at Navy Federal Credit Union, USA

Email:

Address:21000 West Ten Mile Road, Southfield, MI, 48075-1058, Southfield, Michigan, United States, 48075