Enterprise Leaders Chart Path Through AI's "Year of Agentic" Transformation

 

AWS, SAP, and Accenture executives reveal how businesses are moving beyond experimental AI to production-ready systems that fundamentally reshape operations.

 

The artificial intelligence industry stands at a pivotal moment, transitioning from experimental deployments to production-ready agentic systems that are fundamentally transforming enterprise operations, according to technology leaders speaking at AWS Global Headquarters in Seattle.

 

At a panel discussion titled "Inside the Agentic AI Era", part of AWS's two-day media tour, executives from AWS, Amazon's Artificial General Intelligence Lab, SAP, and Accenture outlined how businesses are navigating the shift from traditional AI to more sophisticated agentic systems—AI that can independently perform tasks, make decisions, and problem-solve without constant human intervention.

 

"What agents can do, which historical software couldn't do, is problem solve in a way that is unprecedented using the power of intelligence," said Vishal Sharma, vice president of Information at AWS AGI. "This is the year of agentic, and next year I think it's also going to be agentic."

 

 

The distinction between traditional AI and agentic systems represents more than mere rebranding.

 

Whilst conventional AI operates on a request-response model, agentic AI can intelligently adapt across complex, end-to-end workflows, making autonomous decisions at various points along the way.
 

 

Real-World Transformations

The practical applications are already delivering measurable results.

 

Rahul Pathak, vice president of Data and AI Go-to-Market at AWS, cited an insurance customer using agentic systems to revolutionise healthcare payment processing.

 

 

By better understanding customer policies and making more accurate prepayment decisions, the company is significantly reducing the need for clawbacks—a major source of customer dissatisfaction.

 

Within Amazon itself, AI has been used to upgrade Java code from older SDKs to current versions, saving 4,500 developer years and £250 million in capital costs through more efficient code, according to Pathak.

 

Financial services are seeing dramatic improvements as well. NatWest Bank has modernised its fraud detection capabilities, freeing up analyst time by 40% and reducing fraud detection costs by 25%, according to Andy Tay, Global Cloud First lead at Accenture.

 

 

Ted Way, vice president and Chief Product Officer at SAP, emphasised four critical components of agentic AI: planning capabilities, evaluation and reflection mechanisms, tool usage, and multi-agent collaboration.

 

"Because agents are powered by large language models, they can collaborate with each other and the human is part of the conversation," he explained.

 

 

 

The Foundation: Data and Infrastructure

Yet the panellists were unanimous that successful AI implementation requires more than just deploying the latest models.

 

Tay stressed the importance of establishing what Accenture calls a "cloud-powered digital core"—a foundational infrastructure that integrates disparate data sources and provides the substrate upon which AI can effectively operate.

 

 

 

"At the moment, if you look at enterprises, there's a siloed nature of data that is sprawled all over," Tay said during the Seattle event. "Getting that established and getting a modern data framework and modern data foundation built is critical."

 

The question of data governance takes on new dimensions in an agentic world. Whilst these systems can break down silos to enable unprecedented collaboration across sales, marketing, supply chain, and HR functions, they must respect existing security boundaries and role-based access controls.

 

"All of the ideas that we have around governance and control and observability—who's doing what with what data, for what purpose—that all really applies in the agentic AI world," Pathak said.

 

He highlighted observability as particularly critical, since each run through an AI system can take different paths based on its goals and reasoning.

 

 

The Human Element

Technology alone won't drive transformation, the panel agreed. Successful implementation requires careful attention to change management and workforce development.

 

Pathak emphasised the need for both top-down leadership support and bottom-up innovation, creating space for employees to develop new work habits.

 

Accenture has retrained approximately 500,000 employees to be AI-literate, according to Tay, applying the same transformation approaches to itself that it recommends to clients.

 

SAP's Way noted the importance of identifying champions and influencers within organisations—early adopters who see results and create a network effect throughout the rest of the company.

 

"Trust is a big thing," Tay added. "There are those who are leading the way, those who trust, and those who are like, 'who's accountable if the AI goes wrong?'"
 

 

Model Selection and Sovereignty

The notion of a single, monolithic AI model has given way to a more nuanced approach emphasising fit-for-purpose models optimised for specific tasks.

 

AWS's Bedrock platform enables customers to select from multiple model families—including Amazon's Nova models, Anthropic's Claude, Meta's Llama, and others—and switch between them with minimal code changes.

 

Sharma cited Trellix, an enterprise security company, as an example.

 

The firm uses Nova models for high-volume security event triage, achieving comparable accuracy at lower cost and higher throughput, whilst employing Anthropic models for generating detailed remediation plans.

 

The discussion also touched on regulatory challenges, particularly in Europe.

 

AWS is launching a European Sovereign Cloud at year's end—a separate entity staffed entirely by EU citizens and nationals, designed to meet stringent EU data sovereignty requirements whilst still providing access to the full range of AWS services and partner capabilities.

 

"It's not a very easy decision for us to create a separate entity," said Ruba Borno, vice president of AWS Specialists and Partners, who moderated the panel. "But we were seeing some customers, because of the requirements placed on them by their regulatory environments, were not able to innovate very quickly."

 

 

Looking Ahead

Despite the rapid pace of change, the panellists identified clear priorities for continued development. Sharma highlighted three areas: improving orchestration between agents, developing better memory capabilities, and above all, ensuring reliability.

 

"Agents are really about reliability, reliability, reliability," he said. "You can't just deploy these systems and walk away from them. They have some degree of agency, so there's often some evolution you need to do."

 

The infrastructure investments being made—including substantial spending on data centres in regions like the Middle East—reflect confidence that demand for AI capabilities will continue to grow.

 

When questioned about whether value creation has kept pace with infrastructure spending, Pathak pointed to dramatic efficiency gains already being realised.

 

Moody's has reduced the time required to generate custom research reports from a week to an hour—a 150-fold improvement, he noted.

 

As enterprises move beyond experimental pilots to production deployments at scale, the message from these industry leaders at AWS's Seattle headquarters is clear: the agentic AI transformation is not coming—it's already here.

 

The question for businesses is no longer whether to adopt these technologies, but how quickly they can build the foundations necessary to compete in an increasingly AI-driven marketplace.