Companies looking to expand from GenAI pilots to agent AI systems need to rethink processes, governance, observability and accountability, rather than treating AI as another automation layer, technology leaders said at ETCIO Annual Conclave 2026. As AI evolves from assisting users to taking autonomous actions, speakers said the next challenge for enterprises is to balance speed, reliability, and operational control while producing measurable business outcomes.
This discussion took place at the ETCIO Annual Conference 2026 session titled “From GenAI to Agentic AI: What it really takes to move from PoC to business impact,” moderated by ETCIO Editor Shantheri Mallaya.
Mukul Jain, CTO of Axis Max Life Insurance, said organizations are increasingly moving away from AI-assisted workflows to systems that enable action and orchestration. However, he cautioned that incremental progress with strong governance and human oversight is needed in customer-facing and regulatory sectors such as insurance.
He shared an example of a multi-agent system that handles customer policy queries. In this system, one AI model drafts the response and another AI model validates the output before delivering it to the customer.
“Human engagement is not a weakness; it is an operating model for this transition process,” Jain said.
He added that companies need to define clear boundaries around areas where autonomous systems can operate independently and where human review remains essential.
Viral Davda, CIO at BSE, said organizations need to clearly differentiate between workflow automation and true agent AI. According to him, AI implementation should start with measurable KPIs and clearly defined business outcomes before expanding further.
He cited how BSE’s AI-driven listing compliance platform has reduced the processing timeline for listing applications from 30-45 days to nearly 1-3 days.
“When you involve decision-making and have measurable outcomes associated with it, you enter the world of agent systems,” Davda says.
Davda also emphasized that companies need to redesign their governance frameworks for autonomous AI, as existing controls built for traditional software systems are insufficient.
Himanshu Pant, CDO at Adani Group, said organizations cannot scale agent AI on top of broken workflows and fragmented data systems. He argued that companies must first fix their underlying processes and establish a strong data backbone before introducing an autonomous decision-making layer.
“If the process isn’t right, AI will only accelerate the mistakes,” Pant says.
Pant added that accountability and trust will be key concerns as AI agents move closer to management and decision-making roles within enterprises.
Bharani Subramaniam, CTO of India and Middle East at Thoughtworks, said many companies now describe coordinated workflows as agent AI, even though they are deterministic automation systems. According to him, true agent AI emerges when systems are researching, optimizing, or solving uncertain problems where the steps are not predefined.
“Most of today’s so-called agent systems are actually glorified workflows,” Subramaniam says.
He also emphasized the importance of observability, reversibility, and machine governance. He said companies need to design systems that can detect, audit and reverse incorrect AI decisions when failures occur.
The panel agreed that companies should stop treating AI deployments as isolated experiments. Instead, scaling agent AI requires process redesign, measurable KPIs, richer tooling, data quality, governance frameworks, and operational resiliency.
This session concluded that the future of enterprise AI lies in autonomous systems capable of responsible decision-making, rather than standalone co-pilots. However, leaders emphasized that AI at scale will only succeed if companies build trust, observability, and governance directly into their operating models.
