Enjoyed this article? See more similar articles in ![]()
![]()
Pega Gen AI Cookbook - Recipes ![]()
![]()
series
Agent Debate Pattern in Pega + LangGraph: Every Decision, Justified
What Is the Agent Debate Pattern?
In agentic AI, the Agent Debate pattern replaces a single model making a single call with multiple specialised agents that each argue a different perspective. A Judge synthesises their views into a final, reasoned verdict. Applied to Pega, every case that enters Pega comes with a full AI debate attached — three agents, one verdict, complete transparency.
Without Agent Debate
| Case worker manually analyses the application to reach a decision | |
| Decisions are made by a person — rationale is not always documented in the case | |
| One case worker evaluates the application based on their own judgement and experience | |
| Documentation covers the workflow steps, not the reasoning behind the decision | |
| Every application requires human review regardless of complexity or clarity |
With Agent Debate + Pega
| Advocate, Risk Agent, and Judge each contribute a documented perspective | |
| All three agent opinions attached to the case as evidence from day one | |
| Compliance teams can read exactly why the AI recommended what it did | |
| Arguments are weighed across multiple perspectives before a verdict is reached | |
| If the Judge declines, the A2A call is never made — Pega only receives cases that have passed the debate |
A borderline application triggers the debate. The Advocate highlights stable employment and repayment history. The Risk Agent flags an elevated DTI ratio. The Judge recommends conditional approval at $60k — DTI drops to 39%, full rationale documented. The user confirms. Pega creates the case with all three arguments attached as evidence.
Demo: Click here
Enjoyed this article? See more similar articles in ![]()
![]()
Pega Cookbook - Recipes ![]()
![]()
series