GenAI in application design: does it make a difference? A field experiment using Pega’s Blueprint

From hype to evidence: what we set out to test

Generative AI is becoming part of the architect’s toolkit, but its real value in application design is still taking shape.
To move beyond assumptions, Atos ran a controlled field experiment using Pega Blueprint, comparing GenAI-only, human-only, and mixed teams working on the same design challenge.

What is Pega Blueprint?

Pega Blueprint is Pega’s GenAI-powered design environment for rapidly shaping the foundations of a workflow application. Rather than starting with blank requirements, teams can describe the business problem, select or generate a starting template, and collaboratively define case types, stages, steps, personas, data objects, and integration needs.

The result is a structured application blueprint that can be reviewed with stakeholders and, where relevant, imported into Pega Platform as a starting point for delivery.

For this experiment, Blueprint was not treated as a magic solution, but as a design accelerator.

Its value lies in making assumptions visible earlier, broadening the set of process options, and giving business and IT stakeholders a shared artefact to challenge, refine, and improve.

Speed is easy, quality is harder

The results from the experiment are nuanced. GenAI proved exceptionally fast, producing a complete design within minutes after receiving the final inputs. Yet speed alone did not translate into the best outcome. Human-only teams brought depth and contextual understanding but struggled with time pressure and completeness. The strongest results came from mixed teams, where GenAI and human expertise were combined.

What explains this difference is how each contributes. GenAI excels at breadth, identifying secondary processes and edge cases that humans often overlook. Human architects add structure, interpretation, and refinement. Together, they produce designs that are both more complete and more coherent.

The experiment makes clear that we are moving away from designing fixed, deterministic “happy flows” toward designing systems that can handle variability through orchestration. Where human teams focused mainly on a linear core process, GenAI and especially mixed teams naturally expanded the design to include many more edge cases and alternative paths, reflecting real-world complexity.

This signals a deeper shift: instead of prescribing one optimal process, architects, Business Analists and IT Consultants increasingly define the rules, boundaries, and capabilities within which AI agents can dynamically combine existing workflows, data, and services.
As a result, variability is no longer treated as an exception but as the norm, and the role of architecture shifts from process design to enabling intelligent decision-making across many possible scenarios.

From processes to orchestration

This shift is not just theoretical.In practice, it is already visible in how the GenAI-driven and mixed teams consistently identified more secondary processes and edge cases than the human-only teams.

What we are seeing here is the early form of orchestration. Instead of designing a limited set of predefined process steps, GenAI effectively explores a much wider space of possible scenarios by combining existing rules, flows, and system capabilities. Human architects then validate and structure these into a coherent whole.

In other words, what used to be non-deterministic decision-making in the heads of people is increasingly supported by systems that can orchestrate thousands of deterministic building blocks. The architectural question is therefore shifting: not whether processes should be deterministic, but where the boundary between deterministic execution and adaptive decision-making should sit.

Rethinking the ‘happy flow’

As AI agents gain access to more context and more system capabilities, variability is no longer the exception.Variability becomes the norm. The so-called “unhappy flow” increasingly defines real-world operations.

The focus moves away from designing one optimal process, toward defining the rules, boundaries, and capabilities within which many possible outcomes can be orchestrated dynamically.

Three ways to handle exceptions

In practice, three models start to emerge. (1)Fully deterministic processes are efficient but rigid. (2)Human-driven exception handling is flexible but hard to scale. AI agents introduce a third option: (3)acting as an intelligent exception engine, analyzing full context and proposing tailored decisions, while involving humans only when needed.

The rising importance of decision accountability

At the same time, the role of people evolves. As knowledge becomes more accessible through AI, judgment becomes the differentiator. The ability to interpret context, to make trade-offs, and to take responsibility for decisions—decision accountability—becomes increasingly important.

What this means for consultants and organizations

It is worth noting that the experiment was limited in scale, so the findings are indicative rather than definitive. However, the direction is clear. GenAI does not replace architects or decision-makers. It amplifies them and reshapes how systems, processes, and organizations are designed.

The real takeaway

The future is not about AI alone. It is about how effectively we orchestrate AI, deterministic systems, and human judgment into meaningful outcomes.

Want to see how this works in practice? The full whitepaper explores the experiment, the results, and what this shift means for application design and beyond.