Pega

Agents reach outside systems and other agents through documented rule types. The AI Agent rule builds agents that execute workflows and integrate with third-party agents over MCP and A2A, and a Connect MCP rule points an agent or workflow at an external MCP server so its tools are used during case processing.

Sourcepega.com/insights/articles/introducing-pega-infinity-25-agentic-platform-enterprise-transformationread 2026-09-18

The orchestration layer is Pega Agentic Process Fabric: it analyzes the context of each task and connects it with the best-suited agent automatically, drawing on the full network of registered agents, and coordinates agents, applications, systems and data to get work done. Infinity 26, generally available since 14 July 2026, lets developers connect and orchestrate Pega and third-party AI agents inside governed workflows, and agents are placed as steps within multi-step case lifecycles rather than run as single tasks.

Sourcepega.com/technology/agentic-process-fabricread 2026-09-18

Pega Knowledge Buddy grounds answers in the customer's own content. It ingests existing knowledge bases, connects to existing content storage or builds libraries in Pega Knowledge Management, and analyzes each new piece of content as it is added so the library stays current without retraining. Answers reach employee-facing Pega applications and collaboration channels, and through REST APIs and prebuilt widgets, with transparent attribution to the source content, and access rights control who can see what.

Sourcepega.com/products/genai-knowledge-buddyread 2026-09-18

A person is brought in when a decision needs one. The agentic assignment agent reaches an employee or customer when additional data or approvals are needed to complete a request and pulls them in by email, chat or telephony without manual outreach.

Sourcepega.com/about/news/press-releases/pega-powers-ai-agents-reliably-drive-mission-critical-workread 2026-09-18

The Pega Trust Center publishes its attestations with the documents behind them: a Pega Cloud SOC 2 Type 2 report and SOC 1, a corporate ISO 27001 certificate, ISO 27017 and 27018, a Pega Cloud ISO 42001 certificate for AI management, a HITRUST certification letter, a PCI DSS attestation of compliance, a FedRAMP authorization package, a DoD IL4 provisional authorization, IRAP, C5 and Cyber Essentials Plus.

Pega Cloud for Government holds FedRAMP High Authority to Operate, announced by Pega on 19 March 2025. The customer-facing controls are client-based access control rules that define where and how customer data is stored and accessed, and access-rights controls over who can reach knowledge content.

Sourcepega.com/trustread 2026-09-18

Agent Tracer records every agent interaction, step, request and response, so what an agent did can be inspected step by step, and Pega's orchestration audits agent interactions across the process lifecycle. The GenAI layer is observable in its own right, with usage tracking of request and token volume and audit trails for compliance, and Agentic Process Fabric shows every agent in one gateway. The behavior on the record is the agent's.

Sourcepega.com/insights/articles/introducing-pega-infinity-25-agentic-platform-enterprise-transformationread 2026-09-18

Within a session, conversation state carries as Pega's agents converse with customers and employees in self-service and assisted channels, and every agent working a case is given the full context of that case. The case, though, is the business record itself: deleting what the agent drew on would delete the work item, so it is the application's data model rather than a memory the agent keeps. No memory layer with its own scope and lifetime, separate from the case record, is documented, and nothing describes memories being reviewed, edited, deleted or scoped by user, team or workspace.

Sourcepega.com/technology/agentic-process-fabricread 2026-09-18

Infinity 26 is available through Pega Cloud or in the client's own managed cloud, which is a named customer environment option. Pega Cloud for Government runs on AWS GovCloud for United States public-sector workloads, and the Pega EU Service Boundary gives organizations in Europe control over where their data lives, who can access it and how it is managed.

Sourcepega.com/about/news/press-releases/pega-infinity-26-now-available-deliver-predictable-outcomes-predictableread 2026-09-18

Pega ships customer-adoptable packaged assets that each do their own job. The document agent, a pre-built agent delivered with Infinity 26, analyzes complex documents, splitting, categorizing, scoring and triaging them for further processing, with a chat feature for asking questions about PDFs and images.

Pega Customer Engagement Studio is an agent-powered workspace that takes marketers from brief to live personalized campaigns, and Pega's packaged applications, among them Customer Service, Customer Decision Hub and Sales Automation, arrive with ready-made workflows the buyer adopts and extends.

Sourcepega.com/about/news/press-releases/pega-powers-ai-agents-reliably-drive-mission-critical-workread 2026-09-18

Work reaches Pega's agents without a person asking for it. An agent is placed as a step in a case workflow and runs when the case reaches that step: Pega states that at each workflow step it can invoke agents to automate manual work, the Agent Step hands judgment-based work such as verifying customer documents to an agent, and GenAI rules added as steps in a case lifecycle are invoked during case processing. The workflow reaching the step is the event that wakes the agent; the conversational side runs through self-service chat, email and voice.

Sourcepega.com/about/news/press-releases/pega-powers-ai-agents-reliably-drive-mission-critical-workread 2026-09-18

The choice of model sits with the customer's own architects. Pega describes its GenAI layer as provider agnostic at the model layer, across Azure OpenAI, AWS Bedrock and Google Vertex, and names selecting the right LLM provider, weighing cost, performance and compliance, as a design decision the customer's lead system architect makes. In AI Designer the builder runs GenAI rules with different model settings and compares outputs across models before choosing. That is customer control of the model rather than vendor-side routing.

Sourceacademy.pega.com/node/103571read 2026-09-18

The platform is callable from outside at both design and run time. Since Infinity 26, generally available from 14 July 2026, customers invoke Pega agents and workflows from AI assistants including Claude Cowork, ChatGPT and Office 365 Copilot over MCP, and connect coding tools such as Claude Code, GitHub Copilot and OpenAI Codex to Pega Blueprint, Pega Infinity Studio and Pega Platform over MCP to build applications.

Sourcepega.com/about/news/press-releases/pega-infinity-26-now-available-deliver-predictable-outcomes-predictableread 2026-09-18

Changes are tested before they reach production. Pega uses AI to generate scenario tests for a customer's workflows, including the agent steps inside them, and the tests run directly in Pega or export to third-party testing tools. For the AI components themselves, GenAI rules are prompt-tuned and unit tested in the AI Designer playground to confirm responses meet business expectations before deployment.

Sourcepega.com/insights/articles/introducing-pega-infinity-25-agentic-platform-enterprise-transformationread 2026-09-18

Pega ships its own robotic automation that operates real application interfaces. In Pega Robot Studio, developers interrogate Windows and web applications by dragging a target over the specific control an automation needs, such as the user name field, password field and login button of an application's sign-in screen, and build the automation on the captured controls.

Match rules decide whether a captured control is still available to an automation, Pega states those rules may change when an application or web page updates, and the interrogation form flags any control it can no longer match. Pega Robot Manager runs these robots attended on a user's desktop or unattended, and Robotics Autopilot builds an automation from a plain-English task description.

Sourceacademy.pega.com/node/15241read 2026-09-18