Stop Reasoning So Much, Pega Tells Enterprise AI

Whac-A-Mole is still a fixture at arcades. You get a hammer, moles pop up at random, and you swing at whichever one shows its head.

"Think of the hammer as AI," Deepak Visweswaraiah, Senior Vice President and MD of Pega India, said to AIM Media House. People go looking for anything that resembles a nail, he says, and keep swinging the hammer called AI at it.

That is Visweswaraiah's read on the last two years of enterprise AI adoption. Pega, the enterprise AI and workflow automation company, has built its product and pricing around that read, betting the rest of the industry will follow.

A Whac-A-Mole Problem

Don Schuerman, Pega's Chief Technology Officer, traces the pattern to an earlier hype cycle. Microservices architectures went through the same arc a few years back, he says, when every application seemed destined to be split into thousands of tiny services, until the cost of scaling that many services caught up with the enthusiasm.

Not every application needed a thousand microservices, he says. Maybe it needed 10.

AI is having the same reckoning, Schuerman argues. Large language models, he says, are "not particularly good at doing things that need to be deterministic," where the rules are fixed and need to run the same way every time, because they "introduce randomness where you don't want it." Not everyone in the industry draws that line in the same place, but NVIDIA's enterprise partners have made a similar case recently, arguing industrial AI is moving past copilots and toward autonomous systems built for production rather than open-ended experimentation.

Design Time, Not Runtime

Pega's answer is architectural. Push AI reasoning to the design phase, and let runtime run the same way every time.

"I'm going to use AI at design time to make that as efficient and good and as customer-centric as possible," Schuerman says, "and then I want to just run it repeatedly and consistently at runtime." Document handling and summarization, in his view, are jobs AI still does well even after that design work is finished.

This thinking is now built into the product. Pega's Blueprint AI, first released in 2024 as an early design tool, has grown since into something Schuerman calls core to nearly every delivery the company does. With Pega Infinity 26, released in July 2026, that same engine has been folded into Pega's build environment, Infinity Studio, spanning the full path from idea to design to build.

Pricing as the Guardrail

Pega has priced its product around that same argument.

The company moved away from user-based pricing nearly a decade ago in favor of charging per case, meaning per dispute resolved or account opened. With Infinity 26, that logic now extends to AI itself, a flat fee per resolved caseregardless of how much AI ran behind the scenes, instead of metering tokens. Pega also introduced a token cost calculator alongside the new pricing, letting enterprises estimate what they would have paid under a consumption model before switching over.

The rest of the enterprise software market is arriving at a version of the same idea. SAP has started pricing its AI push around outcomes rather than raw model size, and Oracle has rolled out outcome-based pricing for some of its agentic capabilities, charging per candidate screened rather than per token consumed. The shared worry underneath all of it, as one recent piece on enterprise AI usage metrics put it, is that companies keep confusing usage with value.

"Just because AI is there doesn't mean that every decision, every query has to actually go to an LLM," Visweswaraiah says. "You don't need that level of reasoning." Simpler, cheaper methods handle most queries, in his telling, and Pega reserves the model for cases it considers to actually require reasoning.

Where the Judgment Still Lives

Both executives draw a boundary around what they think AI will never touch. "AI doesn't know what the business is trying to achieve," Schuerman said to AIM Media House.

AI, he adds, will never understand what it's like to be a customer who is "really anxious because they've got a charge on their credit card they don't recognize." As AI makes it possible to build more software faster, Schuerman says the scarce resource shifts toward human judgment about what is worth building at all, a dynamic GitHub's leadership has pointed to as well.

That judgment is also what Pega says it now screens for in new hires. Faced with two graduates carrying identical AI skills, Visweswaraiah says he picks whoever shows sharper curiosity and a better instinct for asking the right question, because the specific skills a graduate walks in with will be outdated within a year or two.

Much of that hiring pipeline runs through India, where Visweswaraiah says the company now houses 38% of its headcount and 60% of its engineering, a base Pega has kept investing in even as it leans on the same design-time approach everywhere else.