PegaWorld 2026: The Year Agentic AI Had To Prove Itself

Enterprise AI has entered a more demanding phase, where agentic systems must prove they can deliver predictable outcomes in real business operations. PegaWorld 2026 framed that shift around workflow discipline, cost control, governance, and enterprise readiness.

This episode explores six lessons for leaders scaling AI beyond pilots.

TLDR / At a Glance

• Predictable AI and governed execution
• Outcome based AI cost control
• Closing the strategy to execution gap
• Orchestrating agents through approved workflows
• Legacy modernisation as AI readiness
• Enterprise discipline in AI assisted development

AI agents have had years to impress us. PegaWorld 2026 forces a tougher standard: prove you can run inside complex enterprises without cost surprises, compliance gaps, inconsistent decisions or another layer of fragmented tech. That shift matters if you own regulated operations, customer outcomes, technology risk, or a budget that has to hold up when usage scales from a pilot to millions of interactions.  

We dig into six takeaways that keep agentic AI trustworthy. The big one is predictable AI: do the heavier reasoning upfront when redesigning workflows and operating models, then keep live execution tight by using lighter AI to understand intent, select an approved workflow, and follow it consistently. We also unpack why ambiguity is the real project risk, and how tools like Pega Blueprint aim to turn business intent into build-ready workflow designs that can be governed, reused and audited.  

Cost becomes a board-level conversation when token-based pricing meets long context windows and multi-step processes. We argue for measuring AI economics by outcomes such as cost per completed case, not prompts, tokens or model calls, and explain how deterministic workflows can narrow agent scope to reduce spend and risk. From orchestration and Model Context Protocol through to legacy COBOL modernisation with AWS Transform, we connect the dots between workflow automation, AI governance, and true AI readiness. 

If you care about enterprise AI that lasts, subscribe, share this with a colleague, and leave a review with the one workflow you would redesign first.

#PegaPartner

Transcript

From AI Hype To AI Proof

SPEAKER_00 0:00

Pega World twenty twenty six The year agentic AI had to prove itself. Six key AI insights for business leaders. Las Vegas did not feel like a conversation about AI potential this year. It felt like a conversation about AI proof. At Pega World 2026, the enterprise AI debate clearly moved on. The question was no longer whether AI agents can generate impressive outputs, respond quickly, or demonstrate clever new capabilities. The harder question was whether they can operate inside complex, high-volume, regulated business environments without creating cost surprises, compliance gaps, inconsistent decisions, or another layer of fragmented technology. That is the shift business leaders should pay attention to. The next phase of enterprise AI will not be judged by how impressive the demo looked. It will be judged by what the workflow delivered. For me, Pega World 2026 was about one central idea. Agentic AI will only scale when it is designed into the operating model, governed through workflows, costed against outcomes, and engineered for the realities of enterprise work. Across the keynotes, demonstrations, innovation hub conversations, and product announcements, the message was consistent. Enterprise AI is moving away from isolated pilots and toward governed workflows that can be trusted, measured, and scaled. Here are my six biggest takeaways.

Predictable AI Beats Powerful AI

SPEAKER_00 1:34

Predictable AI is now more important than powerful AI. The strongest message from PEGA World 2026 was that enterprise AI has to become more predictable before it can become truly scalable. That may sound less exciting than the usual AI language, but it is far more important for leaders responsible for regulated operations, customer outcomes, technology risk, and cost control. PEGA's predictable AI architecture is built around a useful distinction. Heavier AI reasoning is used at the design stage, where teams are rethinking workflows, processes, rules, and operating models. When the system is live, lighter AI is then used to understand the user's intent, select the right approved workflow, and follow the process consistently. That matters because many agentic AI approaches ask the AI to keep working out what to do while the work is already happening. That may be acceptable in lower risk scenarios, but it is much harder to justify in banking, insurance, healthcare, public services, customer operations, or compliance-heavy environments where decisions must be auditable, repeatable, and explainable. A customer service agent handling a simple query may have some room for flexibility. An agent supporting a loan approval, claims decision, eligibility assessment, or regulated customer interaction needs something different. It needs clear steps, clear rules, clear accountability, and a reliable path from request to resolution. The practical distinction is that AI does the more creative and analytical work upfront, helping teams shape the process before it goes live, while live execution is focused on recognizing the request, selecting the approved workflow, and carrying it out consistently. As Matt Healy, Senior Director Product Strategy and Marketing put it, the biggest source of project risk is ambiguity. Blueprint helps teams align on the workflow, the requirements, and the outcome before build work begins. That upfront, clarity matters because trusted AI agents should follow approved workflows, not decide the process from scratch every time. Ask whether your agents are operating inside a governed workflow or improvising one. That question separates AI that demos well from AI that can be trusted in production.

Cost Control Moves To The Board

SPEAKER_00 4:08

AI cost control has become a board-level issue. The second major theme was cost. For the past two years, many organizations have treated AI cost as an experimentation issue. Licenses, pilots, test usage, internal sandboxes, and proof-of-concept work were manageable enough because usage was limited and the financial exposure was relatively contained. That changes when agents start operating across thousands or millions of enterprise interactions. Token-based pricing can make costs hard to forecast because organizations are charged based on how much text the AI reads, processes, reasons through, and produces. A simple user request can create a lot of hidden work behind the scenes. For example, a user may ask an AI agent to check a customer case and recommend the next action. Behind that simple request, the AI may review case history, inspect policies, search previous interactions, call another tool, compare options, summarize the answer, and check the response again. While this may be manageable during pilots and controlled experimentation, it becomes much harder to defend once AI agents are running across thousands or millions of interactions, where unpredictable token consumption quickly turns from a technical detail into a financial control issue. Pega's response is to shift the cost conversation away from AI activity and toward completed business work. The point is not how many tokens were consumed, the point is whether the customer request was resolved, the case was completed, the claim was handled, or the order change was processed. Alan Treffler captured this distinction clearly when he said agents, like tokens, are input measures, they're not output measures. There is also a deeper technical point here. The discussion made clear that some of the biggest AI cost risks appear when agents carry long and growing context across multi-step processes. The more an agent has to remember, revisit, and reason through, the harder cost becomes to forecast. By using deterministic workflows to hold the overall process context, each agent can be given a narrower, step-specific task rather than being asked to reason through the entire workflow from beginning to end. That matters because smaller context windows do not just reduce cost exposure. They also reduce the risk of slippage, hallucination, and agents going beyond the task they were asked to perform. Taken together, this is the right way to think about AI economics. The value of AI is not measured by the number of prompts, tokens, model calls, agents, or outputs generated. The value is the business outcome achieved. AI cost governance needs to move beyond license management. Leaders should now be asking what work was completed, what it cost per completed case, whether the result was faster, better, cheaper, or more reliable than the current process, and whether the organization can forecast the cost if usage increases tenfold. If those questions cannot be answered clearly, the organization is not yet managing AI as an operating cost. It is still managing AI as an experiment.

Closing The Strategy To Delivery Gap

SPEAKER_00 7:30

The strategy to execution gap is still the real AI barrier. Most organizations do not have a shortage of AI ideas. They have use cases, innovation workshops, strategy decks, pilots, vendor demonstrations, executive ambition, and internal enthusiasm. What they often lack is a reliable path from business intent to production ready systems. That is why PEGA's Solution Designer initiative is important. It points to one of the most practical issues in AI transformation, the gap between what the business wants and what delivery teams can confidently build. This is not just a technical gap, it is an operational gap because someone has to understand the process, capture the intent, align stakeholders, define the workflow, verify the rules, and ensure the design can be built, governed, tested, and deployed. AI does not remove that work. It makes that work more visible. Pega Blueprint AI sits directly in this gap. Its role is to help teams move from idea to build ready workflow design faster while reducing ambiguity earlier in the process. The important point is that Blueprint is not just about faster ideation, it creates a more controlled path where workflows can be designed, reviewed, approved, and reused in a governed way. That matters because rework is rarely caused by technology alone. It is usually caused by unclear requirements, misaligned stakeholders, weak process understanding, or the late discovery of compliance constraints. Take onboarding as an example. A business team may say it wants to use AI to improve onboarding, but that ambition is still too broad to build from. Does it mean faster document checks, better eligibility decisions, fewer handoffs, clearer exception handling, more proactive communication, stronger audit trails, or all of those things together? If those questions are not resolved early, AI only accelerates confusion. As Matt Healy put it, most organizations do not have an AI problem. What they have is an AI execution problem. Before scaling AI, inspect the path from idea to delivery. Identify who owns the workflow design, who validates the business rules, who checks compliance, who decides where AI is allowed to act, and who confirms the system is ready for production. If those answers are unclear, AI will not close the strategy to execution gap. It will expose it. Agentic

Orchestration Over Disconnected Agents

SPEAKER_00 10:05

AI needs orchestration, not more disconnected agents. Another major theme at PegaWorld 2026 was agent orchestration. This matters because the agentic AI market is fragmenting quickly. Organizations are already experimenting with agents built on different platforms, models, and internal systems. Some use OpenAI, some use Claude, some use Gemini, some use AWS, some are building their own agents internally. That creates a new enterprise problem. If every agent needs a custom connection to every business system, AI becomes another integration mess. The result is more tools, more adapters, more inconsistent behavior, more governance overhead, and more things for IT to secure, monitor, and explain. PEGA's support for Model Context Protocol is a response to that problem. In simple terms, Model Context Protocol gives agents a more standard way to connect with business systems and workflows. In Pega's case, the idea is that authorized external agents can discover and execute approved PEGA workflows rather than operating around them. That distinction matters. Imagine a claims agent. In a weak design, the agent might inspect a claim, infer the next step, search for documents, draft a response, and decide when to escalate. That may look efficient, but it also creates risk if the agent is effectively making up the process as it goes. In a stronger design, the agent identifies the customer intent, triggers the approved claims workflow, requests missing documents through a controlled step, checks policy rules, escalates higher risk decisions to a human reviewer, and logs the actions taken. That is a more mature form of automation because the agent is not simply completing tasks, it is operating inside a governed process with clear rules, escalation points, and accountability. This same principle applies across functions. In customer engagement, for example, the issue is not simply whether AI can generate more content or campaign ideas. The issue is whether those actions are governed, relevant, compliant, and connected to the right customer decision at the right time. That is the wider Pega World 2026 argument in miniature. AI becomes valuable when it is embedded inside structured workflows, not when it operates as another disconnected tool. As David Vidoni, CIO at Pega, put it, we've done a lot of work to make sure that whatever applications you're building are secure, adhere to compliance, and most importantly, give you predictable outcomes. For businesses, you cannot have variability where it arbitrarily picks one way or the other depending on when you ran it. The better question for leaders is whether the agent is operating inside clear boundaries, what it can access, what actions it can take, which workflow it must follow, where decisions are logged, when human review is required, and how control is regained if something goes wrong. Legacy

Legacy Modernisation As AI Readiness

SPEAKER_00 13:18

modernization is now an AI readiness issue. One of the most practical announcements at Pega World 2026 was the integration between AWS Transform and Pega Blueprint AI for legacy COBOL modernization. This matters because many enterprise AI strategies eventually run into the same blocker, old systems. Large organizations still depend on legacy platforms that contain critical business rules, customer data, operational logic, and process history. These systems may be stable and reliable, but they are often hard to understand, hard to change, and hard to connect into modern AI-enabled workflows. The real problem is not simply that the technology is old, but that critical business logic, rules, exceptions, and process knowledge are often trapped inside systems that few people fully understand. Traditional modernization can be slow because teams first have to work out what the legacy system actually does. The rules may be buried in decades of COBOL code, old screens, undocumented process knowledge, manual workarounds, and inherited complexity. The AWS and Pega approach changes the conversation. A WS transform helps analyze the legacy COBOL environment and generate documentation that captures the business rules, logic, processes, and data structures. Pega Blueprint AI can then use that output to help design future state, cloud-ready applications and workflows. The important point is that this is not simply about moving old code into a newer environment. That may reduce some infrastructure risk, but it does not automatically make the business more modern, more adaptable, or more ready for AI. The real opportunity is to understand what the legacy system actually does, preserve the rules and logic that still matter, and then redesign the workflow around how the organization needs to operate now. Take an old claims platform as an example. The business may not want to preserve every old screen, every workaround, every manual step, or every exception that is built up over time, but it absolutely needs to preserve the core decision logic, data relationships, compliance requirements, and operational knowledge that make the process work. That is why legacy modernization is no longer just an infrastructure discussion. It is an AI readiness issue. If critical processes are trapped inside systems that cannot easily expose rules, data, decisions, or workflows, then the AI strategy will eventually hit a ceiling. Modernization is not just about reducing technical debt, it is about making the business understandable, adaptable, and ready for AI-enabled execution.

AI Assisted Development With Discipline

SPEAKER_00 16:10

AI assisted development now needs enterprise engineering discipline. The final theme that stood out was the shift from AI assisted coding to AI-assisted enterprise development. That distinction matters. AI coding tools have changed expectations around speed. Developers can now generate code, tests, scripts, summaries, and technical suggestions far faster than before. But enterprise software delivery has never been just about producing code quickly. It is about building systems that are secure, scalable, governed, maintainable, integrated, auditable, and reliable under real operating pressure. That is the space PEGA Infinity Studio is aiming at. The important point is that Pega is not simply saying developers should use AI to write more code. The stronger argument is that AI-assisted development needs to be grounded in architecture, workflow context, reusable patterns, quality controls, testing discipline, and enterprise governance. This is especially important for mission-critical applications. An AI-generated feature may look impressive in a demo, but that is not the real test for enterprise software. The real test is whether it fits the architecture, respects the workflow design, handles exceptions, supports auditability, and remains maintainable once the system is live. Speed is valuable, but speed without engineering discipline simply moves risk further downstream. The same concern came through in the discussion around AI-assisted coding. If an AI tool responds to a small change request by regenerating or recoding too much of the application, the organization may then have to retest far more than intended. A more mature approach uses modular architecture to constrain the change, so AI can update the specific component, rule, or data element that needs attention while leaving the rest of the system stable. That is why Pega Infinity Studio matters in this conversation. It brings the design guidance from Blueprint AI into the build environment, so teams are not starting from a blank page or relying only on generic AI coding suggestions. Developers can still benefit from AI assistance and external coding tools, but within an environment shaped by workflow context, enterprise patterns, governance, and Pega best practice. This connects directly to the wider Pega World 2026 message. Predictable AI is not only about how agents behave once they are running in the business, it is also about how AI-enabled applications are designed, built, tested, changed, and governed before they reach production. Alan Trefler made the risk clear when discussing uncontrolled agent adoption. Enterprises that want consistency of process, consistency of soul, consistency of outcome, he argued, need a better approach than just letting a thousand flowers bloom. That is the core issue for enterprise AI. Value does not come from allowing every team to create its own agents, prompts, workflows, and automations in isolation. Value comes when AI operates inside a design system that reflects the organization's processes, controls, standards, and obligations. This is where Blueprint becomes important. As Karamanton, business director in product engineering at Pega, explained, the latest version of Blueprint lets you get really deep into the application design, exactly how people are going to use that application. You can configure the business rules and the user experience all in Blueprint, well before you start building. Treffler put the wider message more directly. I want you to associate one word with PEGA, predictable. Predictable outcomes, predictable cost. That matters because enterprise AI will not be judged by how impressive it looks in a keynote or demo. It will be judged by whether it can deliver consistent outcomes at known cost inside the operational reality of a business. This is the point leaders should take seriously. AI-supported development should not be judged by speed alone. It should be judged by whether the organization can build faster while still preserving architecture, security, testing discipline, governance, and long-term maintainability. The real test is whether AI helps teams build systems the enterprise can trust. The real test. What the workflow delivers.

Operationalising AI Around Workflows

SPEAKER_00 20:50

PEGA World 2026 was not about AI theater. It was about the operating model required to make agentic AI useful inside serious enterprises. AI value will not come from adding more disconnected agents to already complex environments. It will come from redesigning workflows, governing execution, controlling cost, modernizing legacy systems, engineering reliable applications, and connecting AI directly to measurable business outcomes. That is the difference between using AI and operationalizing AI. The organizations that move fastest will not necessarily be those with the most experiments. They will be the ones that know which workflows matter, how those workflows should change, who owns them, where AI is allowed to act, and what business outcome needs to improve. For business leaders, the practical question is now simple. Where is the one high volume, high friction, high value workflow in your organization that should be redesigned with AI built in from the start? Start there. Because the next phase of enterprise AI will not be judged by what the model can say. It will be judged by what the workflow can deliver. This concludes the article. You can also read this article on my LinkedIn page where I share regular insights on AI, strategy, and emerging technologies.