Prioritizing the transformation of AI architecture into business

Enterprise AI deployments are accelerating, but the pace of change in models, tools and platforms is quietly creating a new and costly trap: costly failures to build replacement AI architectures.

As Google Cloud Next 2026 ensures that agent AI has become the planning method for all of Google LLC’s product portfolios, the tough question for business leaders is not how to build it – it’s how to build it with an early mindset shift. That difference is where many organizations quietly fail, according to Paul Lewis (pictured), chief technology officer at Pythian Services Inc., an AI-owned and consulting firm.

“Whatever you use, your only important non-functional requirement is replacement,” Lewis said. “Make sure that whatever you do, the tool can be changed, the model can be changed, the team can be changed, the technology can be changed, because I guarantee you within weeks – not months, not quarters, not years – it will be different.”

Lewis spoke with CUBE’s John Furrier and Alison Kosik on Google Cloud Next, during an exclusive broadcast on CUBE, SiliconANGLE Media’s live streaming studio. They discussed the evolution of AI architecture, business AI readiness and the gap between production demos and real-world deployment. (* Disclosure below.)

The shift in AI architecture and the productivity gap

The transition from structure to operation is the biggest challenge for this year’s event, Lewis noted. After conducting about 50 customer workshops last year, Pythian found no consistent pattern of AI growth across businesses — responses ranged from “I’ve never heard of it” to organizations reporting multibillion-dollar investments.

“Last year it was about ‘building,'” he said. “Unfortunately, most of them [pilots] never went into production. He ended up doing a lot of internal studies.”

The most common failure mode isn’t technical — it’s the gap between a polished five-minute demo and the months of design and work behind it. Business conflicts, including departmentalization and change management, are not going away because of AI; it has to go with it, explains Lewis. Pythian responded by creating five practice areas – from CTO consulting to managed AI operations – to help customers bridge the divide.

“I can put an agent in production with 70% accuracy,” he said. “You want it in the 90s. That takes rapid change and model change and data source change. They have a lifespan like an application does and that requires a team.”

Here’s the full video interview, part of SiliconANGLE and CUBE’s Google Cloud Next:

(*Disclosure: Pythian sponsored this part of CUBE. Neither Pythian nor other sponsors have editorial control over the content on CUBE or SiliconANGLE.)

Photo: SiliconANGLE