Synopsis: A laid-off Bengaluru engineer opens this essay. I argue that AI-driven job cuts across India and the Philippines’ outsourcing sectors aren’t the inevitable outcome consulting firms claim — their own data shows most AI investment produces no measurable return. The piece names three parties who profit from the inevitability narrative anyway (consultancies, executives, capital markets), then closes on the real claim: every redundancy is a choice made by identifiable people with real alternatives, not a law of physics.
Three years into the job, a software engineer in Bengaluru was told by email that the company no longer needed as many people as it once did. He had recently been made star performer of the year. Nobody disputed the title - it was well earned. The message explained, more or less, that his output could now be produced by systems he had spent those three years training. Nobody has thought to mention that training them was what the work had actually been for.
His case has become commonplace across two industries that spent two decades building themselves around cheap, patient, English-speaking labour. India’s information-technology and outsourcing sector, employing roughly six million people, added close to no net staff across the first nine months of the 2026 financial year, after a decade of steady hiring. Two of its largest firms, Oracle and Tata Consultancy Services, each cut twelve thousand jobs while increasing AI spending. Across the water, the Philippines’ outsourcing trade body has revised its 2028 employment forecast down from two and a half million positions to as few as 1.85 million, and its revenue projection down by close to a fifth, citing the same technology ahead of shifting client habits or sharper competition.
An IMF study of the Philippine workforce found something more precise than the headline figure. Eighty-nine percent of business-process jobs were classified as highly exposed to AI. But sixty-one percent of those exposed jobs were also rated highly complementary, meaning the technology assists the worker in most cases rather than replacing them. Exposure and elimination are not synonyms, whatever the alarming percentage suggests on its own. Still, around one worker in seven in the Philippines faces the risk of displacement. That seventh does not distribute itself evenly. It falls hardest on the entry-level, script-following calls that used to teach a newcomer the trade, the rung an eighteen-year-old climbed onto first, now judged the rung an AI can stand on instead.
A distinct argument has been circulating through corporate seminars and management feeds this year, built from research conducted by the large consulting firms. Bolt AI onto an unreformed process and the gains stay marginal. Tear out the old handoffs, the old approval chains, the old meetings, and rebuild the architecture underneath, and the gains compound. It’s a persuasive account of the moment. It is also, and this bears saying plainly, the exact service sold by the firms whose surveys supply its evidence.
McKinsey’s own 2026 State of AI survey, drawn from close to two thousand respondents across ninety-seven countries, found eighty-eight percent of organisations now use the technology somewhere in the business. Only six percent report it moving earnings by any real margin, and that six percent has barely shifted since the year before, even as spending on autonomous agents has climbed sharply. Two further studies push the figure lower. RAND puts the enterprise AI failure rate above eighty percent. A widely cited MIT project found ninety-five percent of generative-AI deployments left no visible trace on the bottom line. Among the minority who did rebuild their workflow from the ground up, results were markedly better. Most companies didn’t rebuild anything, and most failed regardless of the large language model they bolted on. Effectively, this describes a coin toss dressed as a formula.
Notice, too, the grammar this argument reaches for once it turns from the sector to an individual. What choice are you making? The question positions its reader behind a desk, weighing an option, exercising an agency the sentence assumes everyone in the story possesses. But the person deciding whether to rebuild an operating model sits, almost by definition, near the top of an organisation chart. The redesigned worker sits further down the hierarchy, often several time zones away, absorbing a decision that was never theirs to make and wouldn’t have recognised it as a decision until the email arrived. The second person is no neutral courtesy. It’s one of transformation writing’s oldest devices: place the reader in the role of history’s author, and someone else’s disappearance becomes a choice “we” made together, rather than a cost moved outward from the people who profit by the move to the people who absorb it.
None of this makes the underlying observation false. A chatbot glued onto a broken process rarely produces much of any value. Genuine reinvention does change outcomes, sometimes for the better, and pretending otherwise is just dishonest. But redesign the operating model answers a narrower question than the one an organisation, or a species mid-transition, actually faces. It asks how to pursue the same growth and margin more efficiently. It doesn’t ask whether growth and margin were ever the right units to measure a working life against. Neither does it question why stripping out a handoff between two humans gets filed as progress when that process was often the last point where a mistake was caught before reaching a customer, a patient, or a court filing.
Faster decisions and more accurate ones are not the same claim. By folding them into the same sentence we’re implying that accuracy arrives automatically once a human stops slowing the process down. Accurate against which benchmark? Checked by whom? Answerable to whom when the model is wrong? None of that survives the redesign literature’s confident present tense.
Three actors share this culpability, and the diffusion is not accidental. The consultancies sell a certainty to which they are never exposed: McKinsey’s survey shows most of its clients no better off for having followed the playbook, yet the firm’s fee is unaffected either way. The executive who signs off on a staff cut makes a specific decision that’s reversible right up until the email is sent, then casually borrows the consulting language of inevitability as cover for having made it. Capital markets, for their part, reward headcount reduction on a predictable schedule, independent of whether the AI investment behind it produces anything measurable. So the incentive to redesign survives even when the redesign itself fails.
The engineer in Bengaluru is, in today’s vogue vocabulary, an instance of elimination rather than augmentation, precisely the sorting most routinely used to recommend companies perform faster and more efficiently. He is also, in the version of the story his former employer had no reason to explain, the person who spent three years making the system good enough to file that sorting decision by email. The machine didn’t fall from the sky and replace him. He built the part of it that made his own role dispensable, one annotated call script at a time. His reward was a title the week before that same title became evidence for something else entirely.
He has not, by any account so far, become essential to anything or anyone. Somewhere above the floor on which he used to work, an operating model is being rebuilt around the space he left. The report describing that reconfiguration will most likely record his absence, politely, as an efficiency gained. What it will not say, because no consultancy is ever paid to say it, is that somebody chose this. And a choice, described often enough as inevitable, eventually stops looking like one.
None of which means the cut was owed to physics. Generally speaking, a company facing the same balance sheet issues can keep the handoff, fund the retraining, accept a thinner margin for a season, and most won’t fail because of that. Bengaluru’s engineer could have been moved onto the work his own training data still can’t touch, such as judgment under uncertainty, a client relationship, or the exception the system flags but can’t resolve, at a cost some company director judged not worth absorbing.
Every AI-linked redundancy is the second kind of event dressed as the first: a preference, made by named people with specific alternatives in front of them, then filed under a heading that forecloses the question of whether it had to happen at all.