Enterprise AI is evolving beyond automation into autonomous systems capable of managing complex workflows in real time. As organisations expand AI adoption, issues such as trust, governance, and digital sovereignty are gaining importance, driving greater emphasis on secure data management, compliance, and control. In this interview, Gaurav Agarwal, vice-president, Technology, at IBM India & South Asia, speaks to Sudhir Chowdhary about how the company is enabling enterprises to scale AI responsibly through hybrid infrastructure, governed data, and platforms such as watsonx. Excerpts:
How do AI-led operations differ from traditional automation?
Traditional systems are designed to execute predefined tasks within silos, limiting agility and responsiveness. In contrast, AI agents operate within business workflows, where they can understand context, make decisions, and take action autonomously. In India, where enterprises are scaling rapidly, this shift to workflow-centric operations is critical, enabling organisations to reduce friction, accelerate decision-making, and deliver smarter outcomes through continuous learning and human-AI collaboration.
How can organisations scale AI agents from pilot to production?
Deploying AI agents at scale requires clear intent and disciplined execution. This starts with selecting the right AI agent framework aligned to business goals, while ensuring seamless integration through a hybrid approach with enterprise data and applications. Balance sophistication with simplicity; overly complex systems can slow adoption, while ease of use accelerates value.
Strong data privacy and security controls are essential to build trust, especially as agents interact with sensitive enterprise data. Test rigorously for performance and accuracy, so AI agents can reliably deliver impact as adoption grows.
Finally, AI agents’ orchestration is critical for making them efficient, more collaborative, and easier to scale across the business. Platforms like IBM watsonx enable this by bringing together orchestration, governance, and performance management, helping organisations scale AI agents.
What kind of data architecture enables measurable AI outcomes?
As enterprises scale AI use, data becomes the biggest determinant of success. AI systems need clean, governed and continuously refreshed data, available at speed and scale. Yet in many organisations, data remains fragmented across systems and arrives too late to influence decisions. Achieving real outcomes starts with a unified, real-time data architecture that eliminates silos and makes trusted data accessible. Strong governance, quality and lineage are essential for reliability and compliance. This is why IBM’s recent acquisition of Confluent is significant, enabling real-time data streaming so AI systems can act on continuously updated enterprise data, with built-in lineage, policy enforcement and quality controls.
Which core infrastructure capabilities support agentic AI systems?
The challenge for many organisations today is that their technology architectures were built for a very different era, one that never anticipated today’s hyper-digital, deeply interconnected operating models. These legacy systems lack the modularity and flexibility modern enterprises need, often locking teams into rigid ways of working or long-term vendor dependencies.
To reliably support agentic AI at scale, organisations must adopt a hybrid-by-design approach, where infrastructure is deliberately integrated across on-premises and multi-cloud environments. An IBM study underscores this shift: while 70% of executives say hybrid strategies help optimise cost and performance, only 42% are confident their current infrastructure can handle the data and compute demands of advanced AI models in the near future.
A hybrid cloud strategy unlocks new growth by modernising the foundation of the business, improving agility, scalability, and return on investment. In the race for AI leadership, competitive advantage will belong not to the flashiest AI tool, but to the resilient, invisible architecture that supports it.
Where are AI agents delivering measurable business value today?
AI is already moving from insight to execution. This is visible in our own client zero transformation, where IBM achieved $4.5 billion in productivity gains globally by embedding AI across enterprise workflows. For example, our AskHR agent autonomously resolves 94% of employee queries, reducing support tickets by 75% and lowering operating budget by 40%. In IT operations, AI has reduced support tickets by 56%, with agents resolving nearly 86% of queries.
In finance, AI agents are already transforming trading, compliance, reporting, risk management, and customer service. Their biggest impact is in financial reporting and accounting, where they automate data collection, validation, and disclosure, managing workflows from month-end close to audit readiness while flagging risks and enforcing compliance.
On the factory floor, agentic AI goes beyond defect detection, using real-time data and context to autonomously adjust machines, trigger quality checks, and trace root causes across production processes.
How can enterprises govern AI autonomy while ensuring security and accountability?
Agentic operations must be anchored in digital sovereignty designed to operate across an increasingly complex regulatory environment. Enterprises need data residency, access controls, and policy enforcement embedded into AI systems from the outset. A hybrid, sovereign architecture ensures sensitive data remains within jurisdictional boundaries while enabling AI to scale. Another key element is continuous compliance, aligning AI operations with evolving regulations through real-time monitoring, auditability, and risk controls. By integrating governance and elevating it to a strategic priority, organisations can build trust by ensuring accountability, security, and aligning with business and regulatory expectations.