From cloud to consumer, AI is reshaping business: 'Agents can reset work, not leadership'

AI agents are set to transform how companies work, search and make decisions, but strong governance, clean data and human accountability will be critical when autonomous systems are deployed at scale, business leaders said at a panel moderated by ET's Surabhi Agarwal. Companies should establish governance and processes before scaling the use of agents, Verena Siow, president, Asia Pacific, at SAP, said. "You can automate a process, but you cannot automate leadership," she said.

The central concern around AI is not limited to data security and privacy, experts said, noting that whether the technology was making the right decisions and providing accurate information is equally important.

"AI can be a seductive liar," said Anil Goel, chief technology officer at Nielsen. He stressed the need for clear guidelines and evaluation mechanisms to ensure systems do not drift from their intended purpose.


Nick Schneider, chief executive of Arctic Wolf, said organisations could not yet place blanket trust in one AI agent to oversee another. The risks involved vary according to the use case, data and operating environment, he added. Nigel Vaz, chief executive of Publicis Sapient, compared the challenge in AI agent adoption to the questions companies faced when they began using large human call centre teams with access to customer data. The answer then was to create systems and guardrails defining what each employee was allowed to do, he noted, warning against introducing the technology without redesigning the enterprise around it. Trust will depend on the combination of technology, human oversight, organisational context and clearly defined safeguards, Vaz said. Most companies are currently using AI as a "dimmer switch" to make employees more productive, rather than handing over entire functions to autonomous systems, he said. The first deployments are likely to focus on repetitive work at the edges of business functions. More complex decisions, including credit assessments, trade settlements and insurance adjudication, would require greater caution. Human involvement will remain necessary in high-stakes decisions, Vaz said.

Goel said the extent of human oversight should depend on what was at stake. AI could review reports or past spending with limited risk, but a human would still need to take responsibility for a billion-dollar spending plan, he said.

The panel also discussed the cost of deploying AI and agreed that choosing the right model for each task would become increasingly important as agents are deployed at scale. Sudip Roy, chief technology officer at Adaption Labs, said AI capabilities should be matched to the complexity of the task. "We have (Airbus) A350s, but you would not use one for every journey," he said.

The San Francisco-headquartered Adaption Labs, which Roy founded along with former Cohere colleague Sara Hooker, offers an alternative to the narrative of "bigger models are always better," arguing that the next leap in AI will come from adaptive, specialised systems that learn and improve over time. Vaz said a smaller model could handle simple requests, while more sophisticated systems could be reserved for complex work. AI is also changing how consumers discover products and content. Goel cited a study by Nielsen-owned. Gracenote which found that 49% of Gen Alpha was using AI agents for recommendations instead of search or traditional directories.