Are AI Agents coming for your job? What business leaders say about the future of work

More than 80% of Indian organisations are exploring autonomous AI agents, according to Deloitte's 2025 State of GenAI report. But moving from experimentation to everyday work requires more than an AI model. It requires businesses to decide which tasks an agent can handle, what data it can access and where a human still needs to make the call.

That question came into focus in a recent episode of Mint's All About AI, moderated by Mint Associate Editor Abhishek Singh. The panel featured Aditi Sharma, Vice President of Consumer and Manufacturing Industry Segment at Salesforce India; Sandeep Dewangan, President and Group CIO at Safexpress; and Avnish Anand, Board Member at Orra Fine Jewellery and former CEO and Co-founder of CaratLane. Their experience shows the future of work is not just about job loss or replacement. It is about how tasks are changing and how AI is reshaping what people are expected to decide, analyse, or create.

Starting With Data

For Dewangan, the journey began with a practical challenge. Safexpress had to coordinate information across a large network of vehicles, shipments, branches, and employees. The company built an enterprise data lake and connected its systems. Only then did Safexpress introduce AI for route optimisation, scheduling, and digital assistants to help managers decide. Dewangan believes the first investment should be in information infrastructure.

Sharma describes enterprise architecture in four layers: data context, system of work, agent layer, and system of engagement such as workplace platforms. She highlights the importance of metadata and a common business language. Agents must understand what information means before using it. Before involving AI, companies should map where critical data sits and which decisions depend on it. This sequence provides a clear starting point. Identify key decisions, trace the needed information, and determine where an agent can reliably support or execute the work. “A retailer might use AI for routine price updates once the data is right,” she shares.

AI in retail and Logistics

Anand’s work in jewellery shows this in practice. When a customer orders a custom ring, AI instantly estimates the gold weight and pricing. This reduces errors and turnaround time. Companies cannot automate every task, as “Design still needs humans. It is creative work,” Anand says. In customer service, the same balance appears. “AI flags common complaints and tells us where to retrain staff. Managers know where to intervene,” he says.

Dewangan describes how a digital twin, a virtual model of the Safexpress warehouse, helps managers make smarter decisions. With this system, managers know which truck to unload first, how to assign staff, and how to track cargo. Computer vision technology photographs every truck, detects damage, and automatically sends repair tickets. “It’s faster than manual checks,” he says.

For most businesses, AI does not arrive as a sweeping replacement. It starts with targeted activities like matching invoices in finance, tagging products in retail, or flagging late deliveries in logistics. Anand explains, “It is about speeding up routine tasks so people can focus on disputes, creative campaigns, or safety issues.”

Not Every Decision Can Be Automated

Giving AI more responsibility means defining clear boundaries. Sharma points out that data governance and trust are essential since agents need rules about what information they can use and what actions they can take. Anand’s experience with customer service shows why these boundaries matter. “AI handles common cases well, but unusual complaints need people. There is always a way to transfer to a human,” he says. The level of AI autonomy depends on the business. Dewangan explains, “Marketing can experiment and accept some mistakes. But in logistics, a dispatching error delays urgent deliveries. In support, chatbots handle routine queries, and humans take over for anything unusual.”

Before deploying an AI agent, companies must define the decision it will make, understand the possible consequences of mistakes, and set clear points for human review. Automation should be designed considering error costs, not just because the technology is available.

Managers Will Have Different Jobs

As AI takes on more tasks, management is changing. Sharma expects that by 2030 she will manage teams of both people and AI agents. Managers will need to decide which work belongs to an agent, evaluate its output, and know when human intervention is necessary.

Dewangan expects AI to become, in his words, “orchestrated at scale and embedded everywhere”. He focuses on building systems where intelligence is part of daily decision-making. Dewangan believes leaders should measure AI by the improvements it brings to processes, not by the number of models launched. Anand sees these changes from the customer side. He expects retail to move toward “N=1 personalisation” where businesses respond to each customer’s needs. Analysts’ jobs are shifting too. As AI takes over more data collection and preparation, human analysts will focus on interpretation and problem-solving.

That shift raises new workforce-planning questions. Managers must ask which tasks people should do, which agents should handle, and what new responsibilities will matter as routine work declines. These include overseeing AI systems, managing customer relationships, and developing new services.

What Humans Will Continue To Bring

Sharma points to empathy and compassion, crucial for sensitive customer issues. Dewangan emphasises judgement, especially when rerouting deliveries in emergencies. Anand believes creativity is essential for designing jewellery collections that break with tradition. Professionals should review their daily work, identify repetitive or rule-based activities, and learn AI tools that can speed them up. They must also keep building their judgement, communication, problem-solving, creativity, and people skills.

For leaders, adopting AI is about more than buying the latest tool. It means redesigning how work flows and helping employees take on new roles like troubleshooting AI errors, interpreting AI-generated insights, or training others to use new systems. Above all, leaders must set clear decision boundaries and check whether technology truly improves costs, speed, service, or quality.

The future of work is not arriving in one big leap. Instead, it is coming through many small, tangible changes. The most successful companies know what to automate, what to redesign, and what work is best left to humans.