The enterprise AI conversation is undergoing a fundamental shift. After two years of experimentation with large language models, organisations are beginning to recognise that deploying AI at scale requires far more than selecting the right model. The real challenge lies in embedding AI into business operations while ensuring governance, compliance, measurable outcomes and seamless integration across enterprise systems.
While generative AI has demonstrated remarkable capabilities, many organisations continue to struggle with moving initiatives beyond proof-of-concept stages. According to Sharath H. Keshavamurthy, Director – Solutions Engineering, South Asia, UiPath the next wave of enterprise AI will be defined not by better models but by stronger orchestration and operational discipline.
In this exclusive interaction with Express Computer, Keshavamurthy discusses why enterprises are getting stuck between experimentation and production, why governance must become part of AI design from day one, and how business orchestration is emerging as a critical layer for enterprise transformation.
The industry is still only scratching the surface of AI
Although generative AI has rapidly entered mainstream enterprise discussions, Keshavamurthy believes most organisations are still using it in relatively limited ways. “If you look at what happened over the last one or two years, people started getting a glimpse into what an LLM can do. But we are still barely scratching the surface. People are still using it like a cool search or a chatbot. While some organisations have gone beyond that, not everyone has been able to scale. The larger opportunity is applying agentic AI to transform business processes rather than simply helping individuals complete tasks.”
He argues that enterprises are now entering a more mature phase of adoption, where the focus shifts from exploring technology capabilities to solving concrete business problems.
Why AI pilots struggle to reach production
Many organisations invested heavily in experimentation during the early stages of generative AI adoption. However, those pilots were often isolated from broader enterprise considerations.
According to Keshavamurthy, this disconnect explains why many promising projects fail to scale. “The experimentation phase was really about exploring capabilities. Organisations were not thinking about the enterprise as a whole. There were no guardrails and no integration strategy, and that’s where many organisations got stuck. They explored AI but didn’t know where to fit it into the business.”
Rather than starting with technology demonstrations, he believes enterprises should begin with business processes and identify where AI can create measurable operational improvements alongside automation and existing enterprise systems.
ROI still determines enterprise AI success
Despite the excitement surrounding AI, Keshavamurthy says organisations continue to evaluate investments using familiar business metrics.
Technology alone is never enough; initiatives must improve operational efficiency, business outcomes and measurable KPIs. He avers, “The conversation hasn’t really changed. Even when automation technologies were being adopted, organisations still had to demonstrate ROI. AI simply becomes another capability within that discussion. The focus remains on business benefits, operational impact and measurable outcomes.”
Business orchestration becomes increasingly important
Enterprise processes rarely exist inside a single application. They typically span multiple business systems, APIs, data sources, employees and increasingly AI agents.
Keshavamurthy believes this growing complexity makes orchestration one of the most important capabilities organisations must develop. “Business processes span multiple applications. An AI agent may be able to make a recommendation, but that recommendation still needs to trigger an action. That action could involve a human, an API, another application or an automated workflow. The challenge is connecting all these components so that work flows seamlessly across the organisation.”
Governance must move to the beginning of AI projects
One of the most common mistakes enterprises continue to make is introducing governance only after successful pilots have already been demonstrated.
Security, compliance and data privacy concerns then emerge as major deployment barriers. “Capabilities are often explored first and governance comes later. That becomes a showstopper because security teams may say certain data cannot leave the organisation or cannot be processed in particular environments. Governance needs to become part of solution design from the very beginning rather than something that is added later.”
For regulated industries in particular, embedding compliance into AI architecture early significantly reduces delays when moving towards production.
Human judgement remains indispensable
While agentic AI continues to become more capable, Keshavamurthy does not believe autonomous systems will eliminate human decision-making in enterprise environments.
Instead, AI should automate repetitive reasoning while allowing domain experts to make final business-critical decisions.
He cites claims processing as an example, where AI can analyse documents, identify rejection reasons and prepare recommendations before presenting them to experienced professionals.
“AI agents can ingest data, perform reasoning and prepare recommendations. The human provides the final validation because domain expertise still matters. What previously took one or two hours may now take only five or ten minutes. That’s how AI creates business impact,” he points out.
Keshavamurthy also emphasises that evaluation frameworks are becoming essential for reducing hallucinations and improving reliability. “Evaluations are a critical component whenever you build an AI engine. They help reduce hallucinations and improve decision-making. But when something is business critical, there is still no way around human oversight.”
Building organisations that can continuously absorb AI
Looking ahead, Keshavamurthy believes the biggest priority for enterprise leaders is creating organisations capable of adapting continuously as AI technologies evolve.
Rather than focusing on today’s models alone, enterprises should strengthen the underlying capabilities that allow them to adopt tomorrow’s innovations. “AI can do far more than humans in many scenarios. The challenge for CIOs is building an ecosystem that can consume these capabilities. Think of AI as fuel that already exists. Enterprises need to build an engine that can consume that fuel. Infrastructure, governance and the right people skills become the pillars that allow organisations to keep evolving as new models emerge.”
As enterprise AI moves into its next phase, Keshavamurthy believes competitive advantage will depend less on access to increasingly commoditised models and more on an organisation’s ability to orchestrate people, processes, governance and intelligent automation into a cohesive operating framework. Those that embed these capabilities early are likely to move beyond experimentation and realise sustained business value from AI.