When agentic AI landed in late 2024, there were endless predictions that AI agents would join the workforce one day. All that remained in the realm of science fiction until McKinsey & Company CEO and Global Managing Partner Bob Sternfels made a stunning statement.
Sternfels, during the Consumer Electronics Show 2025, said the firm has close to 65,000 employees, of which 25,000 are AI agents. The statement was provocative, unsettling, and engineered to prompt a rethink of how work is counted. If companies start counting AI agents as employees, what exactly does the word ‘employee’ now mean?
Sternfels made it clear that the goal of these AI agents is not to replace humans, but to boost productivity. This AI-led transformation is part of what the CEO calls the “25-squared” model. Under this model, the company is growing its client-facing roles by about 25%, while cutting non-client-facing jobs by roughly the same amount.
Follow the Money
For decades, enterprises have used automation, internal tools, and decision systems to amplify human productivity. Of late, Indian IT firms have been integrating AI agents into their core operations, affecting headcount growth. The top five Indian IT firms added only 17 employees to their headcount in the first nine months of FY26.
Tech Mahindra CEO and Managing Director Mohit Joshi framed the shift less as a workforce question and more as a pricing and value problem during the company’s Q3 earnings call.
The company is actively rethinking how work is billed. “Today, AI-driven productivity largely shows up as cost savings that get passed on to clients,” Joshi said. “We want to change that by billing separately for human labour and digital labour.”
One proposal under discussion is to link digital labour pricing to token consumption, an idea Tech Mahindra outlined in a white paper with research and advisory firm Forrester. It is an early signal of how Indian IT firms may eventually monetise AI without tying value directly to headcount.
The financial impact is already visible. Improved productivity, especially in fixed price programmes, has allowed the company to grow revenue without adding employees. Total employee strength stood at 1,49,616 in Q3, down over 3,000 compared to the previous quarter, even as revenue per employee continued to rise.
Joshi rejected the notion that this reflects shrinking opportunity, arguing instead that efficiency gains are freeing capacity and expanding margins. This framing reinforces a central point. Tech Mahindra is now treating AI as an ‘employee’; rather, treating it as a form of labour that must be priced, governed, and measured differently.
Agentic Employees?
However, not all Indian IT companies are following suit. In December, LTIMindtree said it had built 1,500 AI-powered “digital employees”, deploying them across finance, infrastructure operations, and customer service.
Radhakrishnan Rajagopalan, Chief Delivery and Technology Officer at software engineering services company Ascendion, argues that calling AI agents ‘employees’ misses the point entirely. “I would use the term extended team as opposed to ‘employees’. The phrase team or companions directly reflects how AI is evolving as a strategic collaborator,” he tells AIM.
Autonomous systems are learning at breakneck speed. However, Rajagopalan says the risk is not that AI will become powerful; it’s that people feel diminished by how that power is described.
“Leaders must create inclusive, psychologically safe environments where people feel valued, not overshadowed. Involving employees in decisions, giving them ownership, and respecting their autonomy will be key to maintaining engagement,” he says.
That framing also reflects how the nature of technical work is changing. “When we say AI agents are companions or extended team, consider them as those who contribute to research and coding; however, the guardrails, judgement and understanding of the impact lies with humans,” he adds.
This distinction is crucial. Building production-grade software still demands accountability that cannot be delegated to software. Rajagopalan says that AI agents can amplify productivity, but then employee expectations will change.
However, Nitish Shrivastava, Chief Technology Officer – Engineering Hyper-productivity at Persistent Systems, believes that AI agents will start entering the workforce. Citing Moltbook, a social network where AI agents are conversing with one another, Shrivastava notes that the capabilities of AI agents are truly remarkable.
“They [agents] have started giving better results, because they are all controlled, they are created in a way that is repeatable. So they are not hallucinating anymore. I think the industry is slowly drifting towards all that,” he says.
Future of Work
Jaspreet Bindra, Co-founder of AI literacy initiative AI&Beyond, takes a more interpretive view of why firms are using the term ‘employee’ at all.
“The term isn’t merely rhetorical. It signals that autonomous AI systems are increasingly embedded into core workflows, not just used as tools on the side,” he notes.
Bindra agrees that this integration is real. Firms are redesigning work so that humans spend more time on judgement, creativity, and client interaction while AI handles analysis and synthesis at scale. At the same time, he warns against letting the metaphor run unchecked.
“Calling AI ‘employees’ can blur lines around accountability, quality and responsibility,” he says. “Only humans can be held legally and ethically accountable for decisions, something regulators, clients, and society will increasingly demand.”
Early deployments already talk in terms of millions of hours saved. The deeper shift, Bindra argues, is how work itself is measured. “This transformation accelerates the transition from work as hours logged to work as impact delivered,” he says. “That has the potential to boost innovation and job satisfaction where humans apply strategic judgement.”
The risks are structural, not theoretical. Entry level tasks that once formed the base of career ladders may shrink, forcing companies to rethink how talent is developed. Errors and hallucinations from autonomous systems can undermine trust if human oversight is eliminated.
Bindra believes this is what AI entering the workforce was always supposed to look like. “AI is becoming a core component of organisational capability,” he says. “What McKinsey’s numbers show is not a dystopian takeover but a symbiotic model where AI and humans collaborate.”
That symbiosis is now moving from consulting decks into delivery organisations. Infosys and Cognizant have both partnered with Cognition, the maker of Devin, an autonomous AI software engineer. The idea is not to replace engineers outright but to embed agentic systems deep into the software development lifecycle.
Devin can plan, execute, and validate end-to-end development tasks, while tools like Windsurf augment engineers in real-time.
Cognizant has said that 30% of its code is already generated with AI, with a target of 50% in the near future. Through the partnership, the company plans to combine autonomous engineering capabilities with enterprise-grade governance, platforms, and scale. Infosys has announced a similar deployment across internal engineering operations and client engagements.
At digital engineering company Encora India, the debate is less about terminology and more about measurability. Amardeep Singh, SVP and BU Head, argues that calling AI agents ‘employees’ is largely a communications choice. What matters is whether they perform repeatable, accountable work that previously required human effort.
“At the end of the day, we track AI the same way we track people, through output quality, cycle time, error rates, and business outcomes,” Singh notes. At Encora, AI is already embedded into delivery pipelines for code review, test generation, documentation, and workflow automation. “These systems don’t replace engineers. They expand their capacity, consistency, and speed.”
He sees clear upsides in treating AI as a visible part of the workforce. Quantifying AI contribution makes capacity planning and pricing more honest, especially when a meaningful share of delivery is AI-accelerated. But he cautions that the label can overpromise. “Many AI systems still work well only within defined boundaries. Calling them ‘employees’ can imply levels of reliability and autonomy that aren’t always there.”
For Singh, the present moment is transitional. AI is not a replacement workforce, but a capability layer woven into specific tasks. “What we’re seeing now is AI as an embedded peer in workflows, not an independent operator,” he explains. “If companies count AI as part of the team, they also need the same discipline around governance, accountability, and measurement.”
Neither firm has yet classified Devin as an employee. But, as agentic systems take ownership of discrete chunks of work, that line may soon blur.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.