AI Layoffs in India: Real or Excuse? I Checked the Data

What happened

Three people made three claims about AI and jobs, and they can’t all be right. The IMF says almost 40% of the world’s jobs are exposed to AI. Nandan Nilekani called this kind of warning “disaster marketing”. And Zoho’s Sridhar Vembu posted that the outlook for software jobs in general is grim.

I’m an auditor: my job is checking what people claim against what they actually filed, and I have no AI product to sell you. I went through the IMF’s own paper, the quarterly headcount of India’s five biggest IT companies and the government’s own job survey.

Start with that 40%. It’s a global average. In rich countries the IMF puts it at about 60, and when the IMF’s own paper looks at India specifically, it’s 26%. Most Indian workers are farmers, craftspeople and manual workers, and AI can’t plough a field or lay bricks, at least for now.

The second word in that headline matters more: exposed. Exposed doesn’t mean replaced. It means AI can touch your tasks. The IMF splits exposed jobs in two. In some, AI helps you do the job: a doctor with an AI scan reader is still a doctor. In others, AI does the job instead of you. Economists call this complementarity.

The World Bank ran that split for South Asia. The jobs where AI could take over with little left for the human came to about 7%, and it named them: call centre agents, secretaries, programmers.

Where the claims meet the filings

If AI were really replacing Indian tech workers, you’d see it in the headcount. So I checked the headcount.

India’s five biggest IT companies are TCS, Infosys, HCLTech, Wipro and Tech Mahindra. At the end of September 2024 they had 15 lakh 37 thousand people. Two years later, 15 lakh 49 thousand. Up by less than 1% in two years.

So Nilekani has a point. There’s no mass firing in these numbers, and four of the five actually added people last quarter. But there’s almost no hiring either. The jobs aren’t being cut. They’re just not being created.

NASSCOM’s own review in February: revenue up 6.1%, headcount up 2.3%. They have a polite name for that gap, non-linearity, meaning the money grows faster than the jobs.

In August, Vembu had already said exactly this: “We have not laid off people, but we are not creating new jobs either.” And: the money that would have gone to new employees is now going to AI and data centre costs.

The real AI job loss in India so far isn’t the email that says you’re fired. It’s the job that never gets posted.

The evidence an auditor should ask for

When a job isn’t posted, someone specific doesn’t get it. The clearest data on who comes from the US. Stanford researchers tracked payroll data for millions of American workers. In the jobs most exposed to AI, workers aged 22 to 25 are now about 19% below where they’d be if they had kept pace, and it happened mostly through less hiring, not layoffs.

In India, the government’s Periodic Labour Force Survey puts overall unemployment at 5%, steady. But for young people in cities aged 15 to 29 it’s 19%, and it has gone up every single month since May. As an auditor I have to be honest about what this does not prove: that survey doesn’t ask anyone “Did AI take your job?”, and youth unemployment rises for a lot of reasons.

So look at hiring by experience level instead. Naukri publishes this every month. In September, hiring for people with 13 to 16 years of experience was up 7%. Freshers up 1%. AI and machine learning roles up 20%. The doors at the bottom are getting narrower.

Level is half the answer. The other half is the kind of work. A team at MIT studied hundreds of company AI projects, and where job cuts happened they named three places: customer support, administrative processing and standardised development tasks, meaning routine code. They added that “The cuts were hitting work that had been previously outsourced,” and literally wrote BPO elimination.

The World Bank found that after ChatGPT arrived, demand for call centre agents in South Asian job ads fell about 24%. The World Economic Forum’s fastest shrinking jobs read the same: data entry clerks, bank tellers, admin assistants, payroll clerks. And Anthropic, the company behind Claude, found that of the Indian tasks it could match to a job, 45% were software tasks, the highest share of any country it measured.

The pattern isn’t tech jobs or white-collar jobs. It’s standardised work, where someone could write down the steps and check the output without you.

What about the AI layoff headlines? The simplest check: what did the company itself say, and does anything else explain the cut? Livspace cut about 1,000 people in February and did say AI, but reports also noted it hadn’t raised new funding for four years. TCS cut about 2% of its staff last year, mostly middle and senior grades, and its CEO said “This is not because of AI.” Zomato shut a support centre in Hyderabad in August and moved the work to outside partners, with no mention of AI at all. Kamal Karanth of Xpheno, who tracks Indian tech hiring, said AI hasn’t really shown up as a visible cause of significant layoffs in 2026.

So who was right? All three, about different numbers. AI can touch a lot of jobs, but for India it’s about a quarter, and touch isn’t take. The big IT firms aren’t firing. And the jobs aren’t being created.

The test to run tonight

The real question isn’t “will I get fired?” It’s “would my job get posted today?” Write down the five tasks you spent the most time on last week. Then ask three questions about each.

1. Could someone write down the steps? If a new joiner could do it from a checklist, it’s standardised work, the support, admin and routine code bucket. If it needs judgment, context or a relationship, it isn’t.
2. If AI did this task, who would catch its mistake? If it’s you, because you know the client, the system or the rules, that’s the side where AI helps you. If anyone could check it, that task is exposed.
3. Is my level still being hired? Look at your own company’s careers page and your team’s last two backfills. Was the person who left replaced? Are there openings at your level and one below?

Now score it. Tasks that are yes on question one and anyone-could-check on question two: hand those to AI yourself before someone else decides to. Tasks where you’re the one who catches the mistake: do more of those, because that’s the part that’s growing. And if the answer to question three is no, start moving now, while you still have a job to learn on.

If your job is a list of steps anyone could check, AI is coming for the steps. If your job is the checking, AI just gives you more to check.