Enjoyed this article? See more similar articles in 
Pega Cookbook - Recipes series
Pega’s GenAI suite has evolved rapidly, and two of its most powerful capabilities — GenAI Connect and Pega GenAI Agent (Agent Steps / Agentic AI) . Both integrate large language models (LLMs) into Pega workflows, but they serve fundamentally different purposes. Choosing the wrong one leads to over-engineering simple tasks or under-powering complex ones. This article cuts through the noise and explains what each capability does, how they differ, and exactly when to use each.
What Is Pega GenAI Connect?
GenAI Connect is a rule-based, prompt-driven LLM integration built directly into Pega’s workflow automation engine. Introduced as part of Pega’s plug-and-play GenAI architecture, it allows low-code developers to author custom prompts, map LLM responses to case fields, and trigger AI-powered content generation at specific points in a case or process flow.
How It Works
You configure a GenAI Connect rule with:
-
A system prompt (defining the AI’s role and behavior)
-
A user prompt (describing the input and expected output format)
-
A chosen LLM model (GPT-4, Claude, Gemini, Amazon Titan, etc.)
-
Response mappings that write the AI output directly into case fields
The rule runs as an automation step in the case lifecycle. When a case reaches that step, Pega calls the LLM, receives the response, and maps it — all without human intervention.
Key Characteristics
-
No Data Pages required — unlike other GenAI features, Connect does not use Data Pages as a source, making it lighter and faster for targeted tasks
-
Deterministic execution — the prompt and response structure are fixed at design time; the rule always runs the same way
-
Structured output — responses can be mapped to specific fields, lists, or data references
-
AI Designer Playground — lets developers experiment with prompt edits and model configurations in real time without saving changes
-
Supported input formats — Text, PDF (via text extraction), Word (DOCX), and Images (with vision-capable models)
Typical Use Cases for GenAI Connect
| Scenario | Example |
|---|---|
| Document summarization | Summarize a complaint letter and populate a “Summary” field |
| Email drafting | Auto-compose a customer status update email based on case data |
| Data extraction | Pull key fields from an uploaded PDF form into structured data |
| Content generation | Generate a personalized fitness plan from user health inputs |
| Documentation lists | Create required document checklists from case type and jurisdiction |
| Translation or reformatting | Reformat unstructured notes into a structured report |
What Is Pega GenAI Agent (Agent Steps / Agentic AI)?
Pega GenAI Agent is a far more sophisticated capability. Introduced as part of Pega Infinity '25 and the broader Pega Predictable AI™ vision, an Agent is an autonomous, goal-driven AI worker embedded directly in the case lifecycle as an “Agent Step.” Unlike Connect (which executes a single, fixed prompt), an Agent can reason, plan, consult multiple data sources and knowledge articles, and decide what to do next — all within governed boundaries.
Agents are the cornerstone of Pega’s Agentic Process Fabric, which orchestrates multiple agents, workflows, and data sources to complete end-to-end work.
How It Works
An Agent Step is configured in the case type with:
-
A prompt source (a case field that drives the agent’s question)
-
A response target (a case field where the agent’s output is written)
-
Sanctioned tools — the specific data sources, APIs, and knowledge articles the agent is allowed to use
-
Confidence thresholds — governing when the agent acts autonomously vs. escalating to a human
-
Human-in-the-loop controls — checkpoints where a human reviews or approves the agent’s reasoning
At runtime, the agent receives the prompt, reasons over the available tools and data, builds a plan, executes it, evaluates the result, and either completes the task or escalates if something looks wrong.
Key Characteristics
-
Multi-source grounding — uses Data Pages, knowledge articles, policy documents, and case history
-
Reasoning and planning — the agent evaluates its output and adjusts, rather than just returning a single response
-
Full auditability — every reasoning step, tool call, and decision is logged via Agent Tracer
-
Low-code configuration in App Studio / AI Studio
-
Human oversight built in — agents can pause, ask for permission, or escalate
-
Agent chaining — one agent can hand off to another specialized agent to complete a sub-task
Typical Use Cases for GenAI Agent
| Scenario | Example |
|---|---|
| Loan pre-assessment | Agent reviews credit history, payment records, and policy knowledge to recommend eligibility |
| Medical claims triage | Agent evaluates documentation, coverage policies, and prior claims to recommend approval or flag for review |
| Expense review | Agent condenses case audit history and policy articles to brief an approver |
| Fraud triage | Agent detects anomalies, launches an investigation workflow, and escalates to a specialist |
| Customer self-service | Self-service agent understands natural language intent, identifies the right workflow, and executes it end-to-end |
| Ad hoc workflow creation | Agent interprets a manager’s instruction (“arrange an external adjuster”) and builds a structured workflow on the fly |
Head-to-Head Comparison
| Dimension | GenAI Connect | GenAI Agent (Agent Step) |
|---|---|---|
| Primary role | Execute a single AI task (prompt → response) | Reason, plan, and complete multi-step work autonomously |
| Complexity | Low — one prompt, one response | High — multi-tool, multi-step reasoning |
| Data sources | Direct input or uploaded documents; no Data Pages | Data Pages, knowledge articles, case history, external APIs |
| Output | Mapped text or structured data in fields | Recommendations, summaries, next-best-actions, executed steps |
| Governance | Prompt is fixed at design time | Governed tools, thresholds, HITL, full audit trail |
| Auditability | Standard case audit | Agent Tracer with full reasoning chain |
| Human involvement | Depends on the requirement | Configurable checkpoints and escalation |
| Suitable for | Targeted content generation or extraction | Complex assessments, recommendations, and decisions |
When to Use GenAI Connect
Use GenAI Connect when:
-
You need a single, well-defined AI action at a specific point in a workflow (e.g., “summarize this document when the case reaches the Review stage”)
-
The task does not require reasoning across multiple data sources — the input is self-contained
-
You want fast, predictable output with a fixed prompt structure
-
You are performing content generation (emails, letters, plans, summaries) rather than making decisions
-
You need to extract and map structured data from uploaded files (PDFs, DOCX, images)
Think of GenAI Connect as a smart automation step — it calls the AI, gets a result, maps it, and moves on. No reasoning, no escalation, no audit trail of thought.
When to Use a GenAI Agent (Agent Step)
Use a GenAI Agent when:
-
You need the AI to consult multiple sources before producing a recommendation (policy docs, case history, credit data, etc.)
-
The task involves judgment, not just generation — the AI must weigh evidence and provide a rationale
-
You need a full audit trail of why the AI reached a conclusion (regulated industries: banking, insurance, healthcare, government)
-
You want human-in-the-loop checkpoints — the agent should pause for approval before taking certain actions
-
The agent’s output will directly drive routing, decisions, or escalations in the case lifecycle
-
You are building toward autonomous customer self-service that resolves issues end-to-end without human agents
-
You plan to chain agents — where one agent’s output becomes the input for a specialist agent
Think of a GenAI Agent as a co-pilot embedded in your case — it does its homework, shows its reasoning, and hands the case worker exactly what they need to decide fast and correctly.
The Third Option: GenAI Coach
It is worth briefly mentioning Pega GenAI Coach, the third conversational GenAI feature. Coach provides contextual guidance to employees at their moment of need — helping them draft communications, understand next steps, or navigate complex policy — but it is conversational and user-facing, not embedded in automated case flow. If your goal is to assist a case worker with real-time, chat-style guidance (rather than automating a step), Coach is the right choice.
Decision Framework: Which Should You Use?
Does the task require reasoning across
multiple data sources or policies?
│
├── NO → Do you need to generate, extract,
│ or reformat content?
│ │
│ └── YES → Use GenAI Connect
│
└── YES → Does it need a full audit trail
and human oversight options?
│
├── YES → Use GenAI Agent (Agent Step)
│
└── NO → Is it a real-time employee
guidance need?
│
└── YES → Use GenAI Coach
Pega gives you a spectrum of GenAI power. GenAI Connect is the right tool for fast, focused, content-oriented tasks where the input is clear and the output is a single field or document. GenAI Agent is the right tool when the AI needs to think, consult enterprise knowledge, provide explainable recommendations, and operate within a governed, auditable framework.
The cleanest way to remember it: if you are generating something, use Connect. If you are deciding something, use an Agent.
Both capabilities share the same LLM ecosystem, the same low-code development experience in App Studio, and Pega’s core commitment to predictable, governed AI — but they occupy very different positions in the architecture of an intelligent enterprise workflow.
Enjoyed this article? See more similar articles in Pega Cookbook - Recipes series