The Gist
-
Agentic AI obstacles. Legacy systems and technical debt are major blockers to agentic AI success, especially in regulated industries.
-
Smart prototyping tools. Blueprint by Pega helps map old systems and build working prototypes. This makes AI transformation easier to implement.
-
Guardrails for AI. Multi-agent frameworks with built-in guardrails keep AI on track, which helps balance innovation with compliance.
LAS VEGAS — The vision of agentic AI that has been dominating discussions lately still requires facing some often steep obstacles. As they currently stand, many agents are relegated to a limited set of data and tasks. They give inconsistent results and have the potential to run off the rails in terms of compliance in highly-regulated industries.
Yet what I saw and heard June 1-3 at PegaWorld 2025 at the MGM Grand Las Vegas offers a window into how the promise of agentic AI may be realized sooner than you think. These results could be more predictable and dependable, even for enterprises in highly regulated industries.
Pega Founder and CEO Alan Trefler opened the show by using the analogy that, rather than throwing the kitchen sink at a single AI that may excel at some tasks and flounder at others, companies need a conductor that is able to use the right AI tool and AI model for the job. This approach spans from initial planning and design through implementation, integration and ongoing improvement. They call this approach predictable AI.
Here are some of the key takeaways from PegaWorld 2025 and the company’s vision for an agentic future.
Table of Contents
Facing Agentic AI Adoption Challenges
All this talk about agentic AI is great, but how can enterprise companies get there with all of their existing infrastructure?
While all of this sounds like what has been missing from the agentic conversation, some of you may have a nagging feeling. What about all the legacy apps? These could include ancient mainframes, decades-old COBOL code, spaghetti code or systems patched together over 20 years or more. The barriers to adoption are real. Research unveiled at the show from Pega and Savanta shows that 68% of IT leaders say legacy systems are blocking modern tech adoption, and 88% worry that technical debt is letting nimbler rivals sprint ahead.
Understanding the Roots of Technical Debt
Being truly successful with AI transformation has its challenges, and simply adding AI isn’t a sustainable solution. It’s likely doing more to contribute to technical debt than removing any. After all, a lot of modernization is really just putting a modern UI on top of old code. Yet true transformation requires something more. It requires a true understanding of the underlying goals, processes and desired outcomes, as well as any compliance or regulatory constraints.
Pega made some big announcements centered around their Blueprint product, which includes allowing the platform to ingest nearly any legacy artifact. These artifacts include video walkthroughs of a product, documentation, UI screenshots, technical files and even Amazon code reviews. This then allows a full understanding of the internal workflows as well as the customer interactions that they enable.
Prototyping with Generative and Agentic AI
Static documentation only conveys part of the story. Blueprint provides working prototypes for multiple channels/devices and outlines the roles and business processes required. This forms a key benefit and differentiator for Pega amongst a sea of potential competitors. It uses agents to redesign business workflows and customer interactions. This includes web, mobile and phone interactions within an organization.
Once the assets are inside, Blueprint’s agentic AI mines them for hidden workflows and surfaces the underlying process logic. It autobuilds the data fields and object relationships you’ll need for a modern app. Which helps fast-track both migration and integration without a single COBOL or mainframe expert needed. Then it spits out a robust prototype with the ability to walk through the application, step-by-step. Pega’s chief product officer did a live demo from the keynote stage that did all of this, and it let him to call a synthetic customer service agent to solve a real-world problem.
For Pega customers, this greatly accelerates their ability to understand legacy systems and refactor them for a more modernized framework. It also provides clear implementation plans. Even for non-Pega customers, this level of fidelity in prototyping and ideation is a game changer. And, unlike many so-called transformation efforts, this is not just “lift and shift.” It’s something that’s able to analyze code reviews, video walkthroughs, documentation and individual roles both within and outside a company.
Related Article: AI in Customer Service and the Evolving Role of Contact Center Agents
Keeping AI on Track
Greater automation will increase efficiency, but how do organizations maintain control in the areas that matter?
Despite the promise of massive increases in efficiency and speed, many agents are still black boxes that generate inconsistent results and solve challenges in ways that are rather opaque. On top of that, newer LLMs actually hallucinate more than older ones. This means that guardrails must be put in place, and Pega’s solution here is to use different types of AI agents and AI approaches depending on the need, the constraints and the level of control required.
Understanding Pega’s Multi-Agent Approach
Built on Pega’s Process Fabric, which was introduced five years ago, agentic AI is now a core part of their offering. It serves as an agentic conductor and incorporates several types of agents that interact at the most optimal times. This is now referred to as Pega Agentic Process Fabric, and its multi-agent approach includes design-time agents that can incorporate a company’s intellectual property (IP) and best practices.
It also includes run-time conversational agents that use semantic models to determine which approach is best and automation agents that operate inside the individual step of workflow, which allows tighter management of key processes and workflows. Finally, it includes optimization agents that guide users to optimal outcomes via best practices and real-time analysis
Agentic Process Fabric intelligently registers every agent, workflow and dataset, whether that’s Pega-native or third-party. It then initiates the right one the moment a customer or employee needs it. All those assets sit in a single, discoverable library ready for on-demand orchestration, and this offers a high degree of predictability since the right type of AI agent is used for each specific task.
Balancing Innovation With Guardrails
To keep overly creative agents from introducing hallucinations or the wrong elements, Pega splits the jobs. Generative “what-if” ideation happens safely at design time, while a stricter semantic AI takes over at run time, following only vetted workflows. The result is innovation upfront and compliance on the back end. Meanwhile, guardrails are in place to eliminate unknown outcomes in production, when it matters to both customers and the business.
Accelerating Real Transformation
Having a plan is only the beginning. How does an organization speed the time to delivery on AI transformation projects?
It’s not enough to address legacy technical debt and have a solid platform for agentic AI adoption. With the pace of innovation and change, companies increasingly need to move more quickly than ever to make all of this happen. Pega Chief Technology Officer Don Schuermann calls this integrated disruption. This goes beyond simply adding a new interface on top of decades-old code. This approach offers a way to make meaningful improvements without forcing a complete overhaul, which helps you gain real value faster and with less risk.
In addition to accelerating documentation and requirements gathering, Pega announced several key elements in Blueprint. These features support AI-powered work with a developer assistant, backlog generation, UX generation and assistance with integration with third-party platforms. This is already having an impact on the real world, with Ben Cuthbert from Vodaphone sharing onstage that their use of Blueprint has resulted in 30% less rework and a 10% efficiency improvement.
Related Article: Are You Ready for Agentic AI Shoppers as Customers?
Key Takeaways: Themes From PegaWorld 2025
The MGM Grand in Las Vegas was filled with ideas, success stories and truly challenging questions for enterprises under increasing pressure to adapt to competitive pressures and rising expectations from the omnichannel consumer. Yet, a few themes emerged at PegaWorld 2025 that offer a promising path forward.
The first theme was that thinking about agentic AI as a single agent that needs to perform any and all tasks is not a realistic approach. Multi-agent scenarios, where each AI agent plays to its strength and is orchestrated according to a business’ guidelines, is a practical answer.
The second theme was the realization that successful AI transformation remains challenging, with legacy infrastructure and applications standing in the way. Here, AI can now help with documentation and rapid prototyping. It can now help humans think through all the potential challenges, opportunities and risks before a line of code is even written.
The third major theme was the use of AI to assist in all aspects of implementation and optimization. This approach offers faster delivery, tools to prioritize work and methods to optimize long-term performance.
As we’ve seen, each year will bring more advancements in AI and automation, as well as more potential challenges to solve. Yet, a plan for how to use agentic AI for real organizational transformation is emerging, and it will be interesting to see what lies in store at PegaWorld 2026.