Insights on enterprise AI from a recent conversation with Alan Trefler, CEO of Pega
In a recent conversation with Alan Trefler, CEO of Pega, he described a moment from a client meeting that caught my attention. A banking executive stopped the usual enterprise software demo and asked for something more interesting. Within minutes, Pega’s Blueprint tool had mapped the operating model for an entirely new flavored-water business – objectives, workflow stages, data connections, the whole thing.
It’s a funny story, but it points to something more important. Most of the AI conversation today focuses on productivity – summaries, copilots, generated code. But what if the bigger opportunity sits one layer higher? Not automating work, but helping organizations design the processes that determine how the work actually gets done.
That question deserves more scrutiny than the typical demo narrative allows.
The Real Bottleneck Isn’t Code. It’s Process Design.
Writing software has never been the hard part. Agreeing on how the business should actually operate usually is. What organizations struggle with is agreeing on how work should actually flow.
What are the stages of a customer onboarding process?
Who owns the decision logic?
Which data sources must be consulted?
What compliance checkpoints must exist?
Where does human judgment remain mandatory?
Those questions typically take months of workshops, PowerPoint diagrams, and consulting fees to answer. Even then, the results are often ambiguous or contested.
This is where the idea behind Pega Blueprint becomes interesting. Instead of asking AI to generate artifacts at the end of the development process, the platform points the model at the front end of organizational design.
Give the system a goal – launch a new product, redesign a customer service process, handle fraud claims – and it proposes the operational structure required to execute that goal: stages, actors, data flows, user experiences, and policy constraints.
That sounds powerful. But it also raises a more fundamental question:
If AI can propose business processes, who ultimately decides what the organization should look like?
AI as Organizational Architect
Much of today’s generative AI excitement centers on worker productivity: writing emails faster, summarizing documents, drafting marketing copy. Those capabilities matter, but they operate at the margins of enterprise performance.
The real leverage sits elsewhere – in organizational architecture.
Every company runs on an invisible layer of decisions: policies, workflows, rules, escalation paths, eligibility criteria, compliance checkpoints. Those structures determine whether the customer experience feels seamless or frustrating, fair or arbitrary, fast or bureaucratic.
Historically, designing those structures required domain expertise, consulting frameworks, and a lot of iteration.
The proposition emerging from Pega’s approach is different: let AI generate candidate operating models.
That reframes the role of generative AI. Instead of acting as a clever assistant, the model becomes a design collaborator for the enterprise itself.
But collaboration is not the same as authority. And that distinction matters.
The Three Layers of Enterprise AI
The company’s framing of AI capabilities offers a useful lens for thinking about where different technologies belong.
First is statistical AI – the machine learning systems that drive next-best-action recommendations, predictive scores, and optimization models. These are mathematically grounded systems that learn patterns from data and produce measurable performance improvements.
Second is generative assistance. This includes summarization, translation, drafting messages, or pre-filling information for employees during workflows. These capabilities speed up human work but rarely determine outcomes themselves.
The third category – where Pega is pushing – is generative process design.
Here, AI proposes how a business should operate. It suggests roles, workflow stages, decision logic, and data dependencies.
That is a far more consequential use of AI than summarizing a call transcript.
And it raises an obvious question: Should organizations trust AI to propose the structure of their operations?
The answer may depend less on the AI itself and more on when and how it is used.
Design-Time AI vs Run-Time AI
One of the more thoughtful elements of the Pega architecture is the separation between design-time creativity and run-time predictability.
The generative model is allowed to explore possibilities during the design phase. It proposes how a process might work. Humans review the design, test edge cases, and refine the logic.
Once the structure is approved, the process becomes governed workflow logic inside Pega’s runtime environment. At that point the system behaves deterministically. Similar inputs produce similar outcomes.
This sequencing attempts to solve one of the most persistent problems in generative AI deployments: unpredictability in production environments.
Many organizations experimenting with agentic systems discover the same issue. AI improvisation may work in a sandbox. In regulated industries – banking, insurance, healthcare – the idea of allowing an AI agent to improvise customer outcomes is considerably less appealing.
So the model becomes a design partner, not an operational decision-maker.
The obvious question is whether this distinction will hold over time.
Governance Is the Real Battleground
Enterprise AI conversations increasingly revolve around trust. Not trust in the abstract, but trust grounded in governance: data lineage, audit trails, explainability, fairness, compliance.
Executives want innovation. They also want to avoid being the next regulatory headline.
Pega’s messaging leans heavily into this concern. The company emphasizes encrypted data flows, clear decision lineage, and the ability to review every step of the process logic.
From an enterprise perspective, that emphasis is rational. The organizations most interested in AI-driven process design – banks, telecoms, insurers, government agencies – operate in environments where explainability is not optional.
But governance introduces its own tension.
If generative AI becomes powerful enough to design sophisticated workflows, organizations may begin to rely on those proposals more heavily. Over time, the line between “AI suggestion” and “AI authority” can blur.
And that leads to an uncomfortable but necessary question:
Will enterprises truly review every AI-generated design – or will they eventually accept them by default?
The Demo Economy of Enterprise Software
There is another dimension to the Blueprint story that deserves attention: the role of AI in enterprise sales.
For decades, enterprise software sales have depended on lengthy discovery cycles and carefully staged demos. The vendor studies the client’s environment, prepares a tailored presentation, and gradually reveals the solution.
Generative design tools collapse that timeline.
If a platform can analyze a company’s website, ingest a few strategic objectives, and generate a plausible process design in minutes, the first meeting becomes dramatically more substantive.
Consulting partners also benefit. They can embed their domain expertise into reusable blueprint templates, combining proprietary methodologies with AI-generated context.
The result is a curious new artifact in enterprise technology: the AI-generated operating model.
And that raises another strategic question:
Does the company that designs the process eventually control the runtime that executes it?
Platform Gravity
Blueprint outputs can theoretically be exported and rebuilt elsewhere. In practice, the easiest path is usually to execute the design within the platform that generated it.
That dynamic is not new. Development frameworks have always created gravitational pull toward their own ecosystems.
But generative process design amplifies the effect. If the system helps define the business logic itself – not just the code – switching platforms becomes more difficult.
In other words, generative AI may become the newest form of platform lock-in.
Whether that proves problematic will depend on how open these systems become and how easily organizations can move their process logic across environments.
The Larger Question
The flavored water demo is memorable because it compresses a complex idea into a single moment: AI helping design how a business operates.
But the real significance lies beyond the demo.
Enterprise AI may be entering a phase where the technology does not merely automate work. It begins to shape the structure of organizations themselves.
That possibility should prompt deeper reflection.
If AI can propose workflows, policies, and operating models:
Who validates those designs?
Who owns the intellectual property of the process?
How do organizations ensure fairness and compliance across AI-generated systems?
And perhaps most importantly – how do leaders maintain strategic judgment when machines increasingly propose the blueprint for how the enterprise should run?
The next phase of enterprise AI will likely hinge on those questions. The technology can already generate content. Soon it may generate operating models.
The winners won’t be the companies that deploy AI the fastest. They’ll be the ones that decide where AI should design – and where it shouldn’t.
