The short version
In the space of a year, Pega went from “GenAI features inside a workflow platform” to an agentic platform — and unlike most of the industry, it did so with a very specific bet: that enterprises don’t want autonomous agents, they want accountable ones. Having brought Pega’s earlier GenAI wave (Blueprint, Constellation-era features) into production work as an early adopter, here is my read on the two releases that matter.
Infinity ‘25: agents on rails
The headline of Infinity ‘25 is what Pega calls Predictable AI™: AI agents that don’t roam free but execute inside workflows, alongside deterministic steps, with governance inherited from the platform. An agent can read documents, draft a response, chase a missing approval — but the case, the audit trail, and the decision authority stay in the workflow.
If you’ve operated enterprise platforms (I run Pega estates for an EU institution), you’ll recognize why this framing wins procurement conversations: the question every architecture board asks about agents is not “how smart?” but “what happens when it’s wrong, and who can see why?” Putting agents inside case types — with agent rules, an agentic process fabric connecting them, and AI-powered document processing as the first workhorse use case — is a direct answer to that question.
The quieter features matter for delivery teams: an AI developer agent that mentors people through the platform in real time, and AI-generated test suites — which, if they hold up, attack the least-loved and most-skipped part of every Pega project I’ve seen.
Blueprint grew up: from scaffolding tool to legacy X-ray
I evaluated Blueprint when it launched and gave Pega product feedback that made it into the product, so I watch this one closely. The ‘25/’26 wave points Blueprint at a much bigger prize than greenfield design: legacy transformation. AI agents analyze what you already have — documents, process diagrams, screen recordings, even source code — and propose the case types, data models and integrations for the replacement application. The newest updates add natural-language workflow editing and auto-generated implementation plans.
My honest practitioner’s caveat stands from my original evaluation: Blueprint accelerates the first 60% — and an experienced architect still owns the 40% where systems succeed or fail (case design under pressure, data boundaries, integration contracts). Knowing which 60% to trust remains the skill. But pointing that acceleration at legacy estates, where the analysis phase burns months, is the right target.
PegaWorld 2026: the walls opened — on Pega’s terms
Two announcements from June define the direction:
MCP support. With Infinity 26, Pega workflows can be discovered and executed by third-party agents — Claude, Gemini, OpenAI and others — through the open Model Context Protocol. Strategically this is the big one: your Pega processes become callable, governed tools in anyone’s agent ecosystem. The enterprise’s mission-critical work stays deterministic and auditable inside Pega; the conversational front door can belong to whoever the customer already talks to.
No token meter. Pega’s Predictable AI architecture shifts heavy AI reasoning to design time, so runtime agents are cheap and fast — and pricing follows: predictable cost, no metered tokens. Anyone who has tried to budget an LLM feature at enterprise scale understands how unusual, and how deliberate, that is.
Alongside these: Infinity Studio, a reimagined AI-first development environment carrying Blueprint’s best practices into day-to-day building, and new packaged agents (an assignment-chasing agent, a document-processing agent) that automate the connective tissue of case work — as industry coverage noted, the theme was accountability over hype.
What I’d adopt first, and what I’d watch
Adopt first: document-processing agents on high-volume intake (the ROI case is immediate and the failure mode is safe — a human reviews); AI-generated tests on any team with thin coverage; Blueprint on the next legacy-analysis phase you’d otherwise staff with workshops.
Watch carefully: agent behavior under change — models and rules evolve, and yesterday’s validated agent runs in today’s application; MCP security posture (authorized third-party agents executing processes is powerful, and the authorization model is now part of your attack surface — I say that as someone who handles CVE advisories for a living); and the skills gap, because “agentic” doesn’t remove the need for people who understand case design, decisioning and governance — it concentrates the value in them.
That last point is, of course, also a personal one. The platform is moving exactly toward the intersection I work at: workflows, decisioning models, and the discipline to run them accountably. If your team is figuring out what this wave means for your Pega estate — that’s a conversation I’d love to have.
Sources: Pega Infinity ‘25 announcement · Pega agentic AI press release · SiliconANGLE on Infinity 26 · Pega MCP/agents release · Blueprint updates · Enterprise Times on PegaWorld 2026.