Premkumar Balasubramanian, CTO, Hitachi Digital Services
As enterprises accelerate their AI adoption journeys, a quiet but fundamental shift is underway—one that goes far beyond traditional automation. Robotic Process Automation (RPA), once seen as the backbone of enterprise efficiency, is increasingly being challenged by AI-native systems that promise adaptability, intelligence, and autonomy at scale.
In this evolving landscape, Premkumar Balasubramanian, Chief Technology Officer at Hitachi Digital Services, offers a grounded perspective on how generative AI, agentic systems, and orchestration frameworks are reshaping enterprise operations. Drawing from real-world deployments across global enterprises, he explains why legacy automation tools are giving way to AI-driven models—and why the leap from automation to autonomy is more complex than current hype suggests.
From replacing brittle RPA workflows with intelligent agents to building enterprise-grade governance through frameworks like R2O2.ai, Balasubramanian outlines what it truly takes to move AI from experimentation into production. In this conversation, he breaks down the practical realities, lessons learned, and strategic decisions enterprises must make as they prepare for an AI-native future.
Q: How is the traditional Robotic Process Automation (RPA) space being disrupted by new technologies?
The traditional RPA landscape is undergoing a significant transformation, driven by emerging technologies that offer greater flexibility and intelligence. Customers are increasingly evaluating alternatives—and in many cases, replacing legacy RPA platforms altogether.
A key driver of this disruption is the integration of generative AI. Traditional RPA tools, which often rely on brittle, UI-dependent automation scripts, tend to break with even minor interface changes. In contrast, AI-enhanced platforms like Microsoft Power Automate with Copilot offer more resilient and adaptive automation capabilities.
This shift is especially evident in real-world applications. In one of our recent engagements, a legacy RPA-based invoice processing system delivered only 65% accuracy. After transitioning to a fully agentic AI solution, accuracy rose to 92%, resulting in a tenfold reduction in cost per invoice and significantly lowering manual exception handling. We’re seeing similar transitions across clients, many of whom are moving toward AI-native automation strategies.
Q: Is the adoption of AI forcing companies to fundamentally change their operating models?
AI is beginning to disrupt enterprise operating models, though we’re not yet seeing full-scale reinvention across the board. Most large enterprises are centralizing AI initiatives within a Center of Excellence or embedding them into platform engineering teams, where capabilities are developed centrally and distributed across the organization.
Over time, I believe every function within the enterprise will evolve into an AI-native operation—where AI is embedded not just in tools, but in decision-making, workflows, and culture. That transformation will take time—likely 12 to 24 months to fully take root. Traditional companies will feel this shift more acutely than agile, product-oriented organizations.
Today, most AI adoption remains technology-led rather than business-driven. Enterprises are replacing legacy automation tools with AI, but have yet to widely reimagine their business processes—often due to constraints like compliance, legacy systems, and supply chain complexity. This gap between capability and transformation is where the next wave of innovation will likely emerge.
Q: What is the current state of agentic AI and its potential for autonomy?
I’m fully onboard with agentic AI—I believe it’s the future. But we need to be realistic about where we are today. Despite advances in models like GPT-5, we’re still far from achieving true enterprise-grade autonomy.
For example, Klarna deployed AI agents to handle customer service tasks, claiming they matched the output of 700 full-time agents. However, reports surfaced suggesting mixed customer feedback, and some roles were later reinstated—highlighting that fully human-less interaction in critical functions may still be premature.
Another major challenge is orchestration. While multi-agent systems can be built, they’re not yet enterprise-grade. Key features like role-based access controls and secure impersonation—where agents act on behalf of users with appropriate safeguards—are still underdeveloped. Without robust governance, security, and access protocols, agentic AI isn’t ready for widespread enterprise production.
Q: What are some key lessons learned from early AI deployments, and how do they compare to the cloud adoption journey?
One major lesson from early AI deployments is that moving from prototype to production takes significantly longer than expected. While demos can be built in four to six weeks, productionizing those solutions—through accuracy tuning, bias mitigation, and observability—often doubles the timeline.
At Hitachi Digital Services, we’ve developed the R2O2.ai framework—Reliable, Responsible, Observable, and Optimal AI—to address this challenge head-on. R2O2.ai provides a structured approach to making AI workloads enterprise-ready, with built-in principles for trust, transparency, and performance. It’s helping our clients move beyond experimentation into scalable, governed AI operations.
Another key insight is the need for tailored governance and guardrails. Most AI platforms offer generic safeguards, but enterprises require controls customized to their specific risk profiles and operational needs. This has introduced a fresh learning curve for many teams.
The early cloud adoption journey offers a cautionary tale. Many enterprises rushed into migration without modernizing workloads or evaluating ROI, leading to increased costs without proportional benefits. In contrast, AI adoption is unfolding with greater scrutiny. Even in these early stages, organizations are asking, “What value will this bring?”—a sign of growing maturity that could help avoid past missteps.
Conclusion
As enterprises navigate the evolving AI landscape, success will depend on more than just adopting the latest technologies. It will require a fundamental rethinking of how organizations operate, govern, and scale AI.
Drawing from his experience, Premkumar Balasubramanian emphasizes a balanced and pragmatic approach—embracing innovation like agentic AI while recognizing the operational and ethical foundations needed to make it enterprise-ready.
The journey from automation to autonomy is well underway. The enterprises that pair bold experimentation with disciplined execution will define the next wave of transformation.
