Enjoyed this article? See more similar articles in 
Pega Gen AI Cookbook - Recipes series
Enterprise Prompt Fabric Pattern: Governing Dynamic AI Instructions with Pega
Introduction
As organisations accelerate AI adoption, the focus often centres on models, knowledge assets, and orchestration layers. However, one of the most influential components shaping AI behaviour is the prompt itself.
Prompts drive critical outcomes—underwriting recommendations, fraud investigations, customer communications, compliance evaluations, and operational decisions. As AI becomes deeply embedded within enterprise workflows, prompts must evolve from simple instructions into governed business assets—requiring traceability, version management, auditability, and lifecycle control.
To address this need, I developed an Enterprise Prompt Fabric proof of concept using Pega, demonstrating how organisations can dynamically select, govern, and manage AI instructions at scale.
Dynamic Prompt Reference
Traditional AI implementations often rely on static prompts, applying the same instruction regardless of business context.
Prompt Fabric introduces dynamic prompt reference, enabling AI agents to adapt intelligently based on situational factors.
This approach is conceptually aligned with Dynamic Class Referencing (DCR) in Pega, where behaviour is resolved at runtime based on context rather than being statically defined. In a similar way, prompts are not hardcoded—they are resolved dynamically based on business data and decision inputs.
For example, within a commercial lending process, the system can dynamically choose between:
CREDIT_RISK_ASSESSMENT_LOWCREDIT_RISK_ASSESSMENT_STANDARDCREDIT_RISK_ASSESSMENT_DEEP
based on inputs such as:
- Credit score
- Debt-to-income ratio
- Loan amount
- Risk indicators
Key Insight:
Just as DCR dynamically determines the appropriate class and behaviour at runtime, Prompt Fabric dynamically determines the appropriate instruction set (prompt) for the AI agent.
User Instructions are dynamically retrieved from Prompt reposistory.
Outcome:
The same AI agent adjusts its reasoning depth and behaviour according to the complexity and risk profile of each case—delivering context-aware and adaptive decisioning.
Prompt Repository (with Historical Traceability)
At the core of this solution is a centralised Prompt Repository, implemented using a Pega data type to persist and manage prompts as enterprise assets.
In addition to managing active prompts, the repository is explicitly designed to retain historical prompts, ensuring that no previously used instruction is lost as prompts evolve over time. This becomes a critical factor in enabling agents to make informed decisions based on both current and prior prompt logic.
Each prompt record includes:
- Prompt ID
- Capability
- Product
- Complexity Level
- Category
- Status
- Effective Date
- Version
Prompt Repository Insights - Business can add new Insights from the Landing Page. Business can drill down to the prompt and review the Prompt Instructions at any time.
Example capabilities include:
- Credit Risk Assessment
- Income Assessment
- Employment Verification
- Customer Risk Segmentation
- Loan Fraud Detection
This repository serves not only as the enterprise source of truth, but also as a historical ledger of prompt evolution, making it easy to retrieve past versions for auditability and traceability purposes.
Prompt Governance
Enterprises already apply rigorous governance to:
- Business rules
- Decision strategies
- Workflows
Prompt Fabric extends these governance principles directly to AI instructions.
Each prompt goes through a structured lifecycle:
- Review
- Approval
- Activation
- Retirement
- Audit
This ensures that AI behaviour is controlled, compliant, and aligned with organisational standards.
Prompt Versioning
Every update to a prompt results in the creation of a new version—for example:
- Version 1.0
- Version 2.0
- Version 3.0
Critically, all historical versions are preserved within the repository. This allows organisations to:
- Continuously improve prompt quality
- Maintain backward traceability
- Revisit prior logic when analysing decisions
Versioning, combined with the repository design, ensures that prompt evolution is both transparent and trackable.
Auditability
To enable full transparency of AI-driven decisions, each interaction records:
- Prompt ID
- Prompt Version
- Timestamp
- Case Associatio
This makes it possible to precisely determine which prompt—and which version—was used to influence any given outcome, supporting regulatory compliance and internal audit requirements.
Why This Matters
Prompt Fabric elevates prompts from hidden implementation details to governed enterprise assets.
By combining:
- Dynamic prompt selection
- Centralised repository with historical retention
- Governance lifecycle
- Version management
- End-to-end auditability
Pega enables organisations to manage AI instructions with the same discipline applied to enterprise decisioning.
The right prompt.
The right version.
The right business context.