Agents powered by Pega GenAI™ can help you optimize Case processing in your organization by providing conversational support and automating tasks. Users of your application can converse with Agents to obtain the information they need and delegate actions that do not require human intervention.
Case and Application Agents
Depending on how you define the scope of your Agent, and where you place it in the UX, Agents can be divided into Case- and Application-level Agents. Case Agents focus on speeding up Case processing by automating manual work, and typically appear in the utilities pane or a tab of a Case.
Application Agents work across multiple channels to handle incoming requests from users and route them into the correct workflows. For more information, see Case and Application Agents.
The following video provides an overview of Agents in Case Management:
Agent Rules
Both Application- and Case-level Agents rely on Agent Rules.
Agent Rules define what information the Agent can access and what actions it can take. For example, you can create an Agent capable of automating repetitive tasks in a Case, so that Case workers can focus on actions that require human intervention.
In addition, you can configure an Agent Rule to include the following functions:
- Call another Agent.
For example, you can call an Agent that summarizes Case details.
- Run an automated action.
For example, you can run an Automation that changes the Case Stage.
- Use an integration with a buddy.
For example, you can call a buddy that provides information about Pega Customer Decision Hub™.
- Start a Case.
For example, when processing an Expense review Case Type, a Case worker might ask an Agent to start a Case to submit a suspicious amount on the expense.
NOTE: Only Constellation applications support starting Cases by Agents. - Retrieve information from a Data Page.
For example, in a Case to review expenses, you can retrieve the list of recent spending.
- Run a User Action.
For example, you can run a Flow Action that automatically adjusts the Service-Level Agreement.
NOTE: Running a Flow Action from the tool Rule is available only in Constellation applications. When you configure the Flow Action, ensure that you use only scalar properties in the View. Otherwise, the tool Rule cannot resolve the properly and has no effect. - Create an Insight.
For example, visualize all outstanding work in a bar chart.
- Display custom greetings and suggestions.
For example, you can prioritize Agent interactions by providing users with a list of common starter questions.
You define which actions an Agent can perform by creating and referencing tool Rules. For more information, see Specifying knowledge and actions for Agents by creating tool Rules and Associating tools with an Agent.
The following figure shows a conversation with an Agent at run time:
Document processing by Agents
To provide meaningful data for Cases without excessive manual actions by application users, Agents can extract information from documents. Agents can use the extracted data in the following ways:
- To populate the Data Model, if the extracted data matches fields that are
available in the Data Model of the Case Type.
For example, in an Expense review Case, if a customer service representative (CSR) scans an invoice for a reported disbursement, the Agent automatically populates the related fields in the Data Model.
- To analyze and respond with specific information from the document.
For example, if a CSR uploads a PDF file with the company's latest disbursements spread across multiple pages, an Agent can find and provide a specific value, such as a suspicious amount that might suggest a fraud. As a result, the CSR saves a substantial amount of time on manually analyzing the document and avoids the risk of errors.
Agents can process documents by default and no additional configuration is required.
The following table shows supported formats across different LLM providers:
| Attachment type/LLM provider | Google Gemini Flash and Google Gemini Pro | Anthropic Claude Sonnet and Haiku | OpenAI GPT |
|---|---|---|---|
| Image | PNG - image/png, JPEG - image/jpeg, WEBP - image/webp, HEIC - image/heic, HEIF - image/heif | PNG - image/png, JPEG - image/jpeg, WEBP - image/webp, GIF -image/gif | PNG - image/png, JPEG - image/jpeg, WEBP - image/webp, Non-animated GIF - image/gif |
| Document | PDF - application/pdf, JavaScript - application/x-javascript, text/javascript, Python - application/x-python, text/x-python, TXT - text/plain, HTML - text/html, CSS - text/css, Markdown - text/md, CSV - text/csv, XML - text/xml, RTF - text/rtf | PDF - application/pdf, DOCX - application/vnd.openxmlformats-officedocument.wordprocessingml.document | PDF- application/pdf |
| Audio | WAV - audio/wav, MP3 - audio/mp3, AIFF - audio/aiff, AAC - audio/aac, OGG Vorbis - audio/ogg, FLAC - audio/flac | Not supported | Not supported |
| Video | MP4 - video/mp4, MPEG - video/mpeg, MOV - video/mov, AVI - video/avi, FLV - video/x-flv, MPG - video/mpg, WEBM - video/webm, WMV - video/wmv, 3GP - video/3gpp | Not supported | Not supported |
For more information, see Model limitations for attachment support by Agents.
The following video provides an overview of document processing by Agents:
Speech to text conversion
For even faster Case processing, an enhanced user experience, and improved accessibility, Agents now convert speech to text. After clicking the microphone icon, you can dictate a prompt that the Agent converts to text, and then responds to in a written text message. The icon indicates that the recording is on by changing color.
Speech is immediately converted into text and the recording is not physically stored on any device or data storage system.
If you created an Agent in Pega Platform version '25, after you update to Pega Platform version '26, the Agent automatically includes the speech to text conversion feature.
The following figure shows a conversation with an Agent, with an option to record a message, with the changed icon during recording highlighted below:
Agent Step in a Case Lifecycle
You can also incorporate Agents directly into your Case Types by using an Agent Step. When a Case reaches an Agent Step, the assigned Agent performs the defined action asynchronously, increasing the level of automation in your Case processing without restricting the user's workflow. During Case processing, an Agent receives a prompt, and then returns a response as an input to the field that you specify. You can then use the field in your Case Type UI or as an input in various calculations and decisions. For example, in the Expense review Case Type, an Agent receives the prompt to summarize actions that took place in the Case so far, and then displays the summary in the Case.
Agent Steps are best suited to when work needs judgment across several paths rather than one fixed instruction. Compared to a GenAI Connect Rule, Agent Step makes decisions and runs the workflow Steps by itself, while a GenAI Connect Rule runs a single action as defined in its configuration. Both Rules are governed by the Case.
For more information, see Automating work by running an Agent.
Agent versus Coach
Both Agent and Coach Rules provide interactive support for run-time users of your application relying on Pega GenAI; however, their options differ.
The following table lists differences between Agent and Coach Rules:
| Agent | Coach |
|---|---|
| Agent can run an automated action, communicate with a Pega Knowledge Agent™, call another Agent, retrieve information from a Data Page, and start a Case. | Coach is designed to answer questions as part of a guided experience, to provide reliable and consistent responses. |
| Users can enter their own questions and also select from predefined questions. | Users can select only from predefined questions. |
| Support for documents upload in questions. | No documents in predefined questions. |
The following list includes similarities between Agent and Coach:
- Identical run-time presentation
- Ability for users to specify system instructions and data sources.
- Inclusion of a starter question and quick-select questions.
To provide solutions in various business scenarios, you can create multiple Agent and Coach Rules and integrate them with the same Case Type. At run time, users can select with which feature they want to interact, as shown in the following figure:
