Cookbook Recipe – Pega Agentic Email

[VALLC (Cedric Vallette)]

Introduction

Pega Agentic Email brings agentic AI into the email channel to help optimize intake and processing of customer requests! It leverages a Pega GenAI Application Agent that will handle multi-turns conversations, apply case knowledge, analyze attachments, integrate with workflows, and manage escalation logic.

Pega Agentic Email is available with 25.1.2 (Pega Platform and Pega Customer Service) in PREVIEW mode only. General Availability is planned with Infinity 26. You can start experimenting, but be aware it comes with limitations.

What it can do

• Identify topic, sentiment and extract meaningful data out from email body and attachments
• Handle the triage process autonomously: Manage back and forth email interactions to help capture additional information when needed, qualify a request, kick-off and progress workflows, or escalate.

Key benefits

• Improve first contact resolution and CSAT with faster, proactive, and personalized email responses.
• Minimize handling costs and agent load with less manual triage and escalations.
• Lightning fast to set up! No model training, no mapping definition for topics or entities, no email reply templates.

Configuring an Agentic Email channel

Let’s give it a try now, it is just a 2 minutes and 3 steps process: HowTo video

Prerequisite: You have GenAI enabled and a working email account configured.

1. Enable Agentic Email capability => Switch on the corresponding toggle
2. Create an AI Agent with Application scope => Reference your targeted case types as tools
3. Create a new Agentic Email Channel => Reference Email Account, App Agent & escalation routing rule

You are now ready to test it!

Notes:

• You may reuse an existing Application Agent (e.g. the one generated by default from Blueprint). Configuring Pega agentic conversational channels (Email, Messaging, Voice) to share the same Agent is actually a good design pattern that further aligns with Pega Center-Out vision.
• Agentic Email tries to auto-map email attachments to corresponding case fields of type Attachment. Therefore, provide meaningful labels for those.
• Same for case types: meaningful description will help auto-map from classification.
• Tailoring the App Agent Instructions, Guardrails or Response style is optional. You may, however, use it to enforce specific logic.
• Email processing is handled using an Email Triage case (Work-Channel-Triage-Email), just like with Email Bot. You may access it from your business case by exposing the Related Cases widget).

Step-by-step guidance

Step 1: Enable Agentic Email capability (toggle)

1. Go to Dev Studio and open the toggles page (Configure>System>Release>Toggles)
2. Search for EnableAgenticEmail and switch it on (it will be off by default)

Step 2: Create an AI Agent

1. Go to App Studio, open the AI Designer Landing Page and create a new Agent. Specify Application as scope.
2. Then, from the Agent configuration page, go to Case Types tab and reference the case types you want the agent to map from identified topics.

Step 3: Create a new Agentic Email Channel

1. Go to the Channels Landing Page, and create a new Agentic Email channel
2. In the channel Overview tab, reference your Email Account.
3. In the channel Behavior tab, reference your Application Agent.
4. In the same page, adjust the routing logic for escalation (manual triage).

Further tailoring

You may empower the underlying app agent for better results, say for instance:

• Add Knowledge Buddies as agent tools to surface documentation about the different target workflows (e.g. for the agent to understand what mandatory documents may be needed to fulfill a specific request and ask them back to sender contact automatically).
• Add additional tools to help the agent perform complementary actions. As an example, Pega Customer Service is providing the OOTB tool CustomerAccountDetailsFromEmail so that Agent can match/verify Contact/Account against corporate SoRs.

• Try different models to assess what works best with your use case. Claude-Sonnet is pretty good – especially for text analysis-, while Gemini-Pro might be a better choice when dealing with attachments (scanned forms, hand-written docs, …).
• Fine-tune Agent instructions, guardrails or response style.
• Design an hybrid approach to optimize email attachments analysis:
  • Let Agentic Email channel perform topic identification and auto-map to business case type and its attachment fields
  • Then leverage a dedicated Connect-GenAI rule within the created business case to analyze those attachment fields, extract data and populate case fields.

Closing notes

I personally find this new Agentic Email capability mindblowing. It already addresses critical gaps that the Pega Email Bot had, such as analyzing diverse types of attachments (complex forms, scans, handwritten documents, etc.) and autonomously asking for missing information to further process a request.

Feel free to use this conversation thread to share your feedback: the use cases you’ve tested, the limitations you’ve seen, and the outcomes you’ve experienced.

Again, do not forget Pega Agentic Email is still IN PREVIEW MODE.

[VikasRaidhan]

This is a well-articulated and practical recipe that does a nice job of showing how agentic capabilities can be applied meaningfully to the email channel. Framing email as an agent‑driven interaction—capable of multi‑turn follow‑ups, attachment analysis, and workflow progression—is a clear step forward from traditional email automation approaches.

One aspect worth emphasizing for teams experimenting with this, especially in preview, is the importance of strong underlying case and escalation design. The quality of outcomes will naturally reflect how clearly workflows, handoffs, and boundaries are defined.

The configuration and setup looks like a breeze. The low‑effort setup is a real strength and makes exploration accessible. As adoption grows, continued focus on governance, predictability, and clarity of agent behavior will be key to scaling this confidently in enterprise environments.

Overall, this is a promising capability, and it’s great to see the conversation shifting toward more outcome‑oriented agentic experiences.

Keen to hear what early experiences others are seeing as they start experimenting with this new capability.

[Kamil_Janeczek]

Interesting to see how quickly it can by developed using Gen AI features. Curious to learn if traditionally used NLP-based Text Analytics (intent/entity models) are still valuable, or we think that Gen AI is full replacing those capabilities.

[VikasRaidhan]

The choice between NLP‑based models and GenAI is a business decision, driven by the nature of the problem.

NLP models are best suited when inputs are structured and deterministic behavior is required. GenAI is most effective when inputs are highly unstructured, variable, and when there is a need to autonomously initiate or process work.

In the Agentic Email channel, workflows act as a guide for the LLM to ensure controlled and predictable outcomes. At the same time, the traditional NLP‑based email channel continues to exist for use cases where determinism and structure are the primary requirements.

Both channels coexist, each optimized for different classes of business problems.