THE TECHNOLOGY INTERVIEW

As President of UST, Manu Gopinath is currently spearheading the company’ s shift toward AI-powered platforms, drawing on his previous sixyear tenure as COO where he built the firm’ s enterprise consulting division.

“ There’ s a lot of enthusiasm as much as there is apprehension,” he says.“ Over the last 18 months, there’ s been a lot of enthusiasm. But now, the reality is large multibillion dollar enterprises are dealing with thousands of different applications in their environment without a lot of integration between the different systems.

“ A lot of patchwork is needed to make it all run together,” Manu continues.“ For AI to work effectively, they need cleaner data and that’ s a big challenge. It is a complex job to get data ready for AI systems.”

This is where the apprehension comes in, once the initial enthusiasm from the AI sales pitch wears off.

“ In a sandbox,” Manu goes on,“ you see a lot more accuracy in terms of how a system works but when you put it into real production data, the accuracy is not that high. This is what people are learning.

“ The solution is a redesign of your workflow, which takes time. So the belief that you can scale rapidly doesn’ t happen that easily. I wouldn’ t say they are insurmountable tasks but it is the operating reality.”

To navigate this gap between sandbox potential and production-level accuracy, forward-thinking organisations are shifting their focus toward hybrid workflows – redefining how human oversight and AI agents coexist on the factory floor and in the back office.

“ Everyone’ s aware of what could be the impact of not having a human in the loop,” adds Manu.“ I see agents coming into an organisation and changing the business in two ways.

“ The human primarily out of the loop is when low-risk, high-volume, repeatable tasks can be looked after by AI. Then, there are some workflows where the human is required to give a trigger or review work.”

For example, an AI agent monitoring factory machinery can predict an imminent malfunction and alert an engineer, proposing a specific fix for human approval before taking action. Manu explains:“ Organisations are looking at the risk profile of the task or the workflow that is getting automated by the agent and then dividing that accordingly by which needs reviewing or more testing, or governance and strict controls.

“ Now over time, we believe the error rates will reduce and so the accuracy or predication rate will improve. This is when tasks may get further automated,

technologymagazine. com 27