There is a very specific sound that happens in a call centre at 2 a.m., and I want you to hear it before we go any further.

It’s not the sound of typing. It’s not the sound of a customer shouting about a delayed refund, though God knows there’s plenty of that too. It’s the sound of forty people on a night shift, in a cold, over-lit office in Nashik or Pune or Gurugram, laughing at something on someone’s phone during a five-minute break, because that laugh is the only thing keeping the shift human. I’ve sat in that room. I still sit in that room some nights, taking calls from people on the other side of the planet who have no idea what time it is where I am, or that I exist as anything other than a voice with slightly-too-correct English and a script I’ve read so many times I could recite it in my sleep, and some nights genuinely do.

That room is exactly the kind of room AI is coming for first. Not because the people in it are replaceable as human beings. Because the job, the specific bundle of scripted, repetitive, rules-based tasks that make up a support call, is exactly the kind of job a language model is embarrassingly good at now. Predictable inputs, predictable outputs, twenty years of recorded transcripts to train on. It’s the easiest kind of intelligence to fake convincingly, and customer support was never going to be the last place that noticed.

So this piece isn’t a hypothetical for me. It’s not a thought experiment I picked because it trends well. It’s Tuesday. It’s the industry I clock into. It’s the thing I think about somewhere around 3 a.m. on a slow shift, staring at a queue of tickets that a chatbot in the next tab could probably answer faster and cheaper than I can, and not lying to myself about that fact.

But here’s what I actually want to talk about, and it’s not the losing-your-job part. Everyone’s talking about that part. Newspapers, LinkedIn thought leaders, your uncle forwarding you an article at 11 p.m. captioned “beta, be careful, dekho what’s happening.” The losing-your-job part is loud, it’s scary, and it is, I promise you, not the part that will actually break you.

The part that breaks you is the Tuesday after. The one where you wake up, and there’s nowhere to be, and nothing that needs you, and you have to figure out who you are when the thing that told you who you were for the last several years is simply, quietly, gone.

That’s the real piece. Stick with me. This is going to be long, properly long, because the topic deserves it, because a five-hundred-word hot take on this is basically malpractice, and because I did the reading so you don’t have to. Grab chai. We have some ground to cover.

Let’s kill the three dumbest arguments in this debate

Before we get into anything useful, I need you to meet three people, because you will meet them at every dinner table, every LinkedIn comment section, and every family WhatsApp group for the next several years, and none of them is worth listening to on their own.

The first is the Denier. The Denier has a cousin who works at a “top MNC” and says AI can’t even write proper code, so relax. The Denier will tell you this exact same automation panic happened with computers, with the internet, with Y2K, and look, we’re all still here, aren’t we. The Denier’s evidence is, invariably, one personal anecdote and a vibe, delivered with the total confidence of a man who has never once been laid off by email.

The second is the Doomer. The Doomer has seen a chart. The chart usually says something like “three hundred million jobs at risk,” and the Doomer will not tell you that this number is a theoretical exposure estimate, meaning “AI could plausibly touch some part of this job,” not a body count of people who actually lost their income. Theoretical exposure and actual disappearance are two very different animals that keep getting mashed together for clicks. The Doomer wants you to feel the exact right amount of dread required to buy their course on “AI-proofing your career,” which, I promise you, will not AI-proof anything, because if it worked that reliably, the man selling it would be too busy using it himself to be selling it to you.

The third one is newer, and honestly more interesting than the first two, because he’s occasionally right. Call him the Washer. The Washer is a company that had a rough quarter, needed to cut headcount for completely ordinary financial reasons, and figured out that “we’re restructuring for the AI era” plays a lot better on an earnings call than “we overhired in 2022 and now we’re paying for it.” Even Sam Altman, whose entire company sells the technology in question, admitted this in public. Speaking at an AI summit in India, he said plainly that there’s real “AI washing” happening, companies blaming AI for layoffs they would have made anyway, sitting right alongside genuine displacement of real jobs by real AI capability, and that even he can’t tell you the exact split. Founder Reports has the full context on this admission.

Here’s where it gets genuinely funny, in the way that only corporate hypocrisy can be funny. Klarna’s CEO spent a solid year as one of the loudest voices for AI-driven headcount cuts, boasting about replacing customer service staff with AI agents. Then customer satisfaction started tanking, badly enough that the company quietly reversed course and started rehiring humans. The CEO himself told Bloomberg the company had “gone too far” and that the AI-only approach produced “lower quality” service than what it replaced. Founder Reports again has the details. Gartner, the research firm that corporate boards actually listen to, is now forecasting that by 2027, roughly half the companies that blamed AI for their headcount reductions will quietly rehire people into similar roles, just under different job titles, so the org chart doesn’t have to admit the mistake out loud.

So the honest answer, the one that fits none of these three people’s scripts, is this: AI-driven job loss is real, it’s accelerating, some of it is being oversold by companies using it as convenient cover, and some of that same cover story will quietly reverse itself in eighteen months once the actual quality gap becomes too embarrassing to hide. All of those things are true simultaneously. Sit with that, because the rest of this piece assumes you can hold several true things in your head at once, which, in my experience, is the one skill LinkedIn discourse is structurally incapable of.

The numbers, because vibes are not data

Okay. Actual numbers. I’m going to give you the global picture first, then bring it home to India, because most of what you’ll read on this topic is written for an American reader, and you, statistically, are probably not one.

Globally, the World Economic Forum’s Future of Jobs Report, which surveyed over a thousand employers representing 14 million workers across 55 countries, projects that by 2030, roughly 92 million existing jobs will be displaced, while 170 million new ones get created, a net gain of 78 million. You can read the full digest on the WEF’s own site. That headline number sounds almost comforting, and the WEF wants it to. But buried in the same report is the number that actually matters to you personally: 39% of the core skills required for today’s jobs are expected to change or become outdated in the next few years, and 63% of employers already cite the resulting skills gap as their single biggest barrier to actually using AI productively. Net job growth at the level of the whole planet is cold comfort if the specific 39% that changes happens to be your 39%.

Goldman Sachs ran its own model and landed on a smaller, more sobering figure: roughly 6 to 7% of jobs displaced over the adoption period, translating to about a one-million-person increase in US unemployment at peak. Fortune’s coverage of the Goldman research lays out the full model. Their own researchers, in the same breath, pointed out that 40% of workers today are employed in occupations that didn’t exist eighty-five years ago, which is the strongest actual argument the Denier has, and it’s a fair one. New categories of work keep appearing. The problem, which Goldman doesn’t spend much time on, is the gap between when the old job disappears and when the new category has enough demand to actually absorb you.

In the US specifically, the tracking firm Challenger, Gray & Christmas has been logging AI-cited layoffs since 2023. The number went from 5,824 in that first year to 54,836 in 2025, and by March 2026, AI had become the single most-cited reason for corporate layoffs that month, ahead of every other cause including plain old “cost cutting.” DemandSage has the full breakdown. As of August 2026, AI-attributed layoffs for the year had already hit roughly 205,000 workers in the US alone, matching the entire prior year’s total in under eight months. Outsource Accelerator’s tracker is worth bookmarking if you want to watch this number climb in real time. The companies on that list aren’t obscure startups you’ve never heard of. They’re Amazon, Meta, Microsoft, Google, IBM, Salesforce, Verizon, Accenture, HP, all named explicitly in a congressional report on the trend. Axis Intelligence has the full company list and methodology. Jack Dorsey cut roughly 40% of Block’s workforce and said plainly that AI has changed what it means to build and run a company. Klarna’s CEO, before his reversal, publicly floated the idea that his white-collar workforce could shrink by a third by 2030.

Now, home turf. India’s technology and BPO sector, the industry I work in, employs about six million people directly and is projected to cross $315 billion in revenue this financial year, according to Nasscom. Policy Circle covers the full picture of India’s outsourcing exposure. That’s the good news, the part that gets quoted in press releases. Here’s the part that doesn’t get quoted as often: India’s top IT firms added a net total of just 17 employees across the first nine months of the last financial year. Seventeen. Not seventeen thousand. Seventeen human beings, net, across companies that employ millions between them. The full story is at Outsource Accelerator. A NITI Aayog report has warned that over 60% of India’s formal-sector jobs could face significant automation risk by 2030, and the World Bank’s own South Asia Development Update found that job listings in occupations most exposed to AI, and least dependent on human judgement, fell by roughly 20% after ChatGPT’s public release. A Storyboard18 analysis found AI can already perform between 20 and 40% of common tech tasks like coding, testing, and documentation, precisely the entry-level work that used to be how Indian graduates got their first foot in the door. Storyboard18’s full analysis is here.

Here’s the specific shape of the damage, because it matters more than the headline percentages do: it isn’t that everyone currently employed is getting fired en masse. It’s that the front door is closing. Analysts are increasingly calling this an “hourglass” labour market, strong demand at the senior, specialised end, strong demand for cheap human labour at the very bottom, and a shrinking, squeezed middle where a fresher used to be able to walk in, learn the ropes, and build an actual career. MIT Technology Review described this as a “low-fire, low-hire dynamic,” where the people who already have jobs mostly keep them, while the pipeline that used to refill the bottom of the funnel quietly seizes up. If you are twenty-two right now, this is the single most important sentence in this entire article: it is not that AI is coming for your job. It’s that AI may be closing the door before you ever get to walk through it.

It’s not just tech support. Let me take you on a tour.

I’ve spent the last two sections talking almost entirely about IT and BPO, because that’s the world I actually live in, and I don’t want to pretend I’m neutral about it. But if you work in literally any other white-collar field and you’re feeling smug right now, please, sit down, because I did the research on your industry too, and it’s not exactly a comfort.

Start with law, because it’s the profession that used to be the poster child for “AI can never replace this, judgement matters too much.” Paralegals and junior legal researchers are now named explicitly on the World Economic Forum’s high-displacement list, a category that felt untouchable as recently as 2024. AI systems now scan legal databases, cross-reference case history, and draft research memos in the time it takes a junior associate to make a cup of tea. Final Round AI’s 2026 breakdown of at-risk roles puts it bluntly: law firms are increasingly replacing entire research teams with software subscriptions. One tracker I came across estimates paralegal salaries declining 10 to 15% even where the jobs themselves survive, while AI-augmented senior associates who know how to direct the tools are seeing pay bumps of 20 to 30% for the exact same underlying skillset that used to be considered mid-tier. Same profession, same degree, opposite trajectory, depending entirely on which side of the tool you’re standing.

Journalism, an industry I care about personally because half my writing life lives adjacent to it, has had one of the ugliest years on record. Newsroom job cuts topped 2,300 across the US and UK in just the first half of 2026, with the broader media and entertainment sector shedding over 7,500 roles. The BBC alone announced cuts of 1,800 to 2,000 jobs, roughly one in ten of its entire workforce, its biggest restructuring since 2011. Claveprep’s tracker of the journalism carnage is genuinely sobering reading if you’re in media. And here’s the part that should make everyone pause before assuming AI equals efficiency equals savings: a Reuters Institute survey of 326 senior media leaders across 51 countries, published in January 2026, found that two-thirds of newsrooms said AI efficiencies had not saved a single job so far. The jobs aren’t disappearing because AI is doing the work brilliantly and freeing people up. A lot of them are disappearing because AI-generated summaries are quietly eating the referral traffic that used to fund the journalism in the first place, which is a completely different, and honestly more depressing, mechanism than the one everyone assumes.

Healthcare offers the most useful contrast of the entire tour, because it shows what happens when a field is judgement-heavy and physically embodied at the same time. Medical coding and the more routine, pattern-matching side of radiology screening are declining. But AI-augmented diagnosticians, health informatics specialists, and clinical AI coordinators are all growing, often commanding a 15 to 25% salary premium over their non-augmented peers, according to AI Magicx’s 2026 disruption report. The lesson isn’t “healthcare is safe.” It’s “the parts of any job that are pure pattern-matching are not safe anywhere, and the parts that require a human to be physically present, accountable, and trusted are safe almost everywhere.” That single sentence, more than any industry-specific stat in this section, is the actual dividing line running through every profession I researched for this piece, not the industry label on your business card.

And then there’s my own other world, the creative one. Writers, comedians, designers, the people I sit with on Wednesday nights working on sets. Production-level creative work, basic graphic design, template-driven copywriting, first-draft video editing, is under real and immediate pressure, exactly as you’d expect. But creative direction, the part of the job that involves taste, timing, knowing which joke actually lands in a room full of actual humans and which one just reads well on paper, has proven stubbornly resistant so far, for a reason that should be obvious to anyone who has ever bombed on stage: a model can generate a thousand jokes a second, and it still cannot feel the specific, humiliating silence that tells you joke number four hundred was the wrong one. That silence is data too. It’s just data no model has access to yet.

Meet Ravi. Meet Ivan. They both trained their replacements.

Statistics are useful for arguments. They are useless for grief. So let me tell you about two people whose stories I came across while researching this piece, because they made the whole thing stop being abstract for me, and I suspect they’ll do the same for you.

Ravi is a junior coder in Bengaluru, three years of experience, and he’d just been named star performer of the year at his company. Weeks later, he got laid off. Not in a meeting. Not even in a phone call. Over email, which included a line that read almost exactly like this: because of AI, we don’t need much resources. Read that sentence again. He was told, in writing, that he was good at his job right up until the moment his job stopped needing to exist, and the company didn’t even feel the need to schedule a call to say so.

Then there’s Ivan, a quality analyst at a call centre in the Philippines, doing work that will sound extremely familiar to anyone in my industry. His job was listening to customer service calls and rating them for quality, correcting the same mistakes over and over, teaching the system what good and bad sounded like. More than seventy QAs at his centre were replaced by AI. Ivan was one of them. What makes his story land differently is what he said afterward: he’d spent months helping improve the very system that ended up taking his job. Outsource Accelerator’s original reporting on both Ravi and Ivan is worth reading in full if this hits close to home the way it did for me.

I want to sit with that phrase for a second, because it’s not just Ivan’s story. It’s the story of an entire generation of BPO and support workers, myself very much included in the professional category if not yet the personal outcome. The same qualities that made our work valuable to global companies twenty years ago, repetition, predictability, doing the same script correctly ten thousand times, are the exact qualities that make it trivial to automate today. We didn’t get replaced despite being good at the job. We got replaced partly because we were good at it. Good enough, consistent enough, to generate the clean transcripts a model needs to learn the pattern in the first place. We were, without quite realising it, doing unpaid data-labelling work for our own eventual successors, one recorded call at a time.

This isn’t unique to tech and BPO, either, if it’s any comfort, which it probably isn’t. A former Head of AI Operations at a company called Pearl was himself displaced by the very technology he now deploys at scale for his employer. His own reflection on it is worth paraphrasing rather than romanticising: he came to understand that his old employer wasn’t transforming, it was optimising, and layoffs give a board clean math and an easy story, while the harder question of what the work should actually become never got asked at all. Fortune published his full account, and it’s one of the more honest pieces of writing on this topic I found in the entire research process. Mass layoffs feel like transformation from the outside. From the inside, they’re usually just cost-cutting wearing a futuristic costume.

The horse in the room

Every conversation about automation eventually arrives at one of two historical comparisons, and I want to walk you through both properly, because one of them is more honest than people think, and the other is less comforting than people think.

Let’s start with horses, because economists genuinely bring this up in serious papers, and it’s a better analogy than it sounds when you first hear it.

In 1983, the Nobel-winning economist Wassily Leontief asked a blunt question: could technological change ever become so total that humans go the way of horses did? Foreign Affairs has the full essay on this question. Here’s what happened to horses, for anyone who hasn’t had this specific nightmare yet: in 1915, there were roughly 22 million working horses in the United States, hauling carts, ploughing fields, pulling carriages through city streets. That was peak horse. Today there are about 2 million, and most of those are pets, sport animals, or ceremonial, not labour. The transition wasn’t instant. It took about 45 years, and by the end of it, 88% of American horses had lost their jobs to the internal combustion engine. Nobody retrained the horses. There was no reskilling programme for carriage horses to become, I don’t know, delivery bicycles. They simply stopped being economically necessary, and the population that depended on human demand for their labour collapsed to match the new, much smaller demand for it.

Every economist who cites this analogy adds the same reassuring caveat immediately afterward: humans, unlike horses, can write laws, invent new wants, retrain, organise, vote, and invent entirely new categories of work that didn’t previously exist. That’s true, and it’s the whole reason economists have historically been right that automation doesn’t produce mass, permanent unemployment across an entire economy. When cheap frozen food and mechanisation gutted farm labour, dropping it from about 40% of the US workforce to roughly 2% over the course of the twentieth century, the economy didn’t collapse into a food-abundant wasteland of the permanently unemployed. It found new things for those people, and especially their children, to do.

But notice what that reassurance actually rests on. It rests on humans being able to retrain and move, not on it being fast, free, or painless for the specific person it happens to. The horse had zero agency in the transition. You have some. The honest question isn’t whether humans as a species will collectively survive AI the way horses didn’t survive the automobile. Economically, almost certainly, we will, because we always have, across every previous wave of this exact panic. The honest question is whether you specifically, this year, with your specific EMI and your specific skill set and your specific runway, have enough time and support to make that retraining work before your savings run out. That is a completely different question, and it’s the one nobody selling you optimism actually bothers to answer.

The weavers who saw it coming a hundred years before the horses did

Before there were horses losing their jobs to engines, there were people, real skilled tradespeople, losing theirs to a much simpler machine, and their story gets told worse than almost any other in this entire genre of essay, so let me try to fix that a little.

You’ve heard the word “Luddite” used as an insult, usually aimed at your relative who still prints out emails. The actual Luddites were nothing like the technophobic cartoon the word has become. They were skilled English textile workers, weavers and stocking-makers, in the early 1800s, and they weren’t protesting machines out of some primal fear of gears and levers. They were protesting a very specific, very rational grievance: mill owners were using new, semi-automated frames to pump out cheap, shoddy cloth, made by unapprenticed, unskilled workers paid a fraction of what a properly trained craftsman commanded, while marketing the result as though it were the same quality product the artisans had spent years learning to make. National Geographic’s history of the movement lays this out clearly: their actual demands included a minimum wage, enforced apprenticeship standards, and a tax to fund pensions for displaced workers. That’s not the manifesto of people afraid of technology. That’s the manifesto of people who could see exactly what was coming and asked, politely at first, for a softer landing.

They didn’t get one. Parliament sided with the factory owners, some Luddites were hanged or transported for machine-breaking, and the transition happened anyway, brutally and fast. The number of British handloom weavers collapsed from roughly 250,000 around 1800 to just 7,000 sixty years later. The word “Luddite” itself got rebranded by the newspapers of the time, most of which were friendly to the mill owners, from “worker with a legitimate grievance” into “irrational technophobe,” and that rebrand stuck so thoroughly that two centuries later, you probably used the word the same way I just described.

I’m telling you this not to depress you further, though I understand if it lands that way, but because the lesson buried in it is genuinely useful and almost nobody extracts it correctly. The Luddites weren’t wrong about the economics. Their wages really did collapse. Their jobs really did disappear on the timeline they feared. Where they miscalculated was tactics, not analysis: smashing the machines slowed nothing, because the economic pressure creating the machines in the first place didn’t go away just because one factory’s frames got destroyed. The lesson for you, sitting here two centuries later reading this on your phone, isn’t “don’t fight it.” It’s “don’t spend your limited energy fighting the machine itself. Spend it building the version of yourself, and ideally the version of your industry’s rules, that survives the transition already in motion.” The weavers who eventually did okay, and some did, were disproportionately the ones who moved early into supervising the new machinery, or into trades the mechanisation hadn’t reached yet, not the ones who held out longest at the loom.

The ATM story everyone tells wrong

You’ve probably heard this one, possibly from someone smug at a dinner party. ATMs got introduced in the 1970s, everyone assumed bank tellers would go extinct, and instead, teller employment actually grew for decades. It’s cited constantly as proof that automation panic is always overblown.

The mechanism is genuinely elegant, so let me give it to you properly. ATMs reduced the number of tellers needed at a typical urban branch from about 21 to around 13. That made each individual branch cheaper to run. Cheaper branches meant banks could profitably open more branches. More branches, even with fewer tellers each, meant total teller employment across the country actually grew, from around 500,000 in 1980 to nearly 600,000 by 2010. CCIA’s writeup on the Bessen research walks through the full mechanism. The tellers who kept their jobs didn’t spend their days counting cash anymore, because the machine did that now. They moved into relationship banking: solving complex problems, selling financial products, doing the parts of the job that required an actual human judgement call. It’s a genuinely lovely story about how automation can shift you up the value chain instead of out of it entirely.

Here’s the part that gets left out at the dinner party, and it changes everything. After 2010, mobile banking arrived, and this time, teller employment actually did start declining, sharply and permanently. The difference wasn’t philosophical. It was mechanical. ATMs automated some of a teller’s tasks, cash handling specifically, and left the judgement-heavy, relationship-heavy remainder of the job intact and even more valuable. Mobile banking automated nearly all of it. There was no remaining “relationship” task left over for the ATM to free the teller up to do, because you can now open an account, apply for a loan, and get basic financial advice from an app on your phone without a human anywhere in the chain.

And here is the sentence that should genuinely worry you more than any headline number in this entire piece: the ATM-to-full-automation transition in banking played out over roughly 40 years, giving the economy, and individual tellers, decades to adjust, retrain, and reorganise around it. The equivalent transition in AI-assisted coding, by multiple analysts’ estimates, is happening in roughly 3 years. Same underlying story. Completely different speed. And speed is precisely the variable that determines whether “the economy eventually creates new jobs” is a comforting fact about humanity in general, or a cruel irrelevance to the specific person whose severance runs out in four months.

Okay, so here’s the part nobody’s actually arguing about

Notice something about everything we’ve covered so far. The Denier, the Doomer, the Washer, the WEF, Goldman Sachs, the horses, the weavers, the ATMs, all of it, every single argument in this entire debate, is fighting over one question: will there be a job for you.

Nobody in that fight is asking the second question, the one that actually determines whether you come out of this okay as a human being. Which is: even if there eventually is a job for you, what happens to you in the meantime? And even when the meantime ends, what did losing the old one cost you that a new pay cheque doesn’t automatically refund?

This is the actual argument of this entire article, so let me say it as plainly as I can. The job loss is the visible problem. It’s the one with headlines, hashtags, and government reports attached to it. The invisible problem, the one nobody’s building a task force for, is what happens to a human being’s sense of self when the structure that organised their days, their identity, and their social worth simply vanishes, sometimes over email, sometimes with two weeks’ notice, and they’re left holding a version of themselves they don’t recognise and don’t know how to rebuild.

That’s the part we’re going to spend the rest of this piece on, because it’s the part that will actually determine whether the next several years break you or just bruise you.

What a job gives you besides money, which is more than you think

There’s a piece of psychology from the 1930s Great Depression that I think about constantly, and I wish it got cited half as often as GDP projections do.

The psychologist Marie Jahoda studied unemployed communities and found something that should be obvious in hindsight but genuinely wasn’t at the time: a job does a lot more for a person than pay them. She identified five specific things employment quietly provides, functions so baked into the structure of having somewhere to be that most people never notice them until they’re gone. A detailed academic summary of her latent functions model is available here. A job gives you a time structure for your day. It gives you social contact outside your immediate family. It connects you to goals and purposes bigger than your own private needs. It hands you a defined status and identity, the thing that answers the question “what do you do” at any social gathering on earth, from a Nashik wedding to a Bengaluru house party. And it forces regular activity, a reason to physically get up and move through the world instead of dissolving slowly into your own mattress.

Take a job away, even with a generous severance package sitting in your account, and all five of those things vanish at once, silently, on the same day. That’s why unemployment consistently correlates with rises in anxiety, depression, and, in the most serious documented cases, elevated suicide risk, particularly among men in cultures with a strong breadwinner identity, who tend to suppress the distress rather than name it out loud, right up until they can’t anymore. Simply Psychology’s overview of the research is worth reading in full, and if any part of this section is landing uncomfortably close to home for you personally right now, that’s genuinely worth a conversation with someone qualified, not just something to push through alone. Researchers studying grief after job loss describe something that reads almost exactly like bereavement: separation distress, difficulty accepting what happened, yearning for the life that existed before, and a genuine confusion about who you are now that the old identity marker is gone.

Here’s the sentence I want you to actually sit with, because it reframes everything: your job was never just a source of income. It was scaffolding. Quiet, invisible, doing five separate jobs of its own that you never had to think about, because you never had to build them yourself. Most people go their entire working lives without realising the scaffolding is there, for the exact same reason nobody thinks about a fridge until the day it stops humming.

This is why “just find another job” is such an incomplete piece of advice, even when it’s well-meant, even when the money side eventually works out fine. Finding another pay cheque solves one of Jahoda’s five functions. It does not automatically solve the other four, and depending on how the job loss happened, it might take a genuinely long time to solve any of them.

The window tribe

I want to give you one more piece of evidence for the identity argument, and it comes from, of all places, corporate Japan, and it’s the single most useful data point in this entire piece for anyone who thinks money alone fixes this problem.

In Japan, there’s a decades-old, semi-official practice called madogiwazoku, which translates roughly to “the window tribe.” Because Japanese corporate culture and labour law make it culturally and legally difficult to simply fire long-serving employees, companies that no longer need someone’s actual output don’t lay them off. They quietly sideline them instead: a desk near the window, minimal responsibilities, occasional emails, full salary, same formal title, right up until retirement. Dror Poleg’s writeup on the phenomenon is a genuinely fascinating read on how this evolved. A 2022 survey of employees at large Japanese firms found that nearly half had a colleague fitting this exact description, someone kept fully paid and fully employed, doing functionally nothing.

Here’s why I’m telling you this. If the theory of the problem were purely financial, purely about the pay cheque, the madogiwazoku should be the happiest workers on the planet. Full salary. Zero performance pressure. Total job security. No risk of the humiliating email Ravi got. And yet the reporting on this phenomenon consistently describes these employees as demoralised, adrift, and quietly miserable, sitting by a window they didn’t choose, doing work that doesn’t matter, watching their professional relevance evaporate in real time while the company keeps paying them not to notice. LatestLY’s explainer on the practice covers the survey data in more detail. Nearly 90% of their own colleagues report that having a madogiwazoku around actively damages team morale, including, one suspects, the morale of the window-sitter himself, watching the whole office quietly agree he’s become decorative.

The lesson is uncomfortable, and it’s exactly the lesson this whole piece is built around. Money without purpose does not fix the hole a job leaves. It just makes the hole quieter, and gives you a more comfortable chair to sit in while you stare into it. If you’re one of the many people currently hoping that a fat severance package, or a future universal basic income, or a generous family safety net, is going to be the thing that gets you through this transition intact, the window tribe is the evidence that you need a second plan, because the first one, on its own, has already been tried, at scale, for decades, and the results were never “everyone’s fine, actually.”

The guilt of the ones who stayed

There’s a quieter casualty in all of this that almost nobody writes about, and it’s the people who didn’t get laid off.

If you’ve watched three rounds of layoffs pass through your office and your name never came up on the list, you’d think that would feel like relief. Often it doesn’t. There’s a documented phenomenon called “layoff survivor syndrome,” and it describes the very real anxiety, guilt, and eroded trust that settles over the people left behind after a round of cuts. You watch a colleague who was, by every visible metric, good at her job, get an email that reads like Ravi’s, and some part of your brain that isn’t especially rational quietly concludes that competence stopped being the variable that determines who stays. From that point on, you’re not working from security. You’re working from the low hum of “I could be next,” which is its own slow erosion, even if you technically still have the job title and the salary that Ravi and Ivan lost.

There’s a related, newer phenomenon that recruiters have started calling “job hugging”: people staying glued to roles they’ve outgrown or actively dislike, not out of loyalty, but out of sheer fear that the market outside the door is worse than the discomfort inside it. It’s a rational response to an irrational market. It’s also, quietly, its own kind of damage, because a person who stops applying for better roles out of fear isn’t advancing, they’re just waiting, and waiting is exhausting in a way that’s hard to explain to anyone who hasn’t done it for a year straight.

I mention this because most articles on this topic split the world neatly into “people who lost their jobs” and “people who are fine,” and that’s simply not how it plays out on the ground. The stress radiates. If you’re currently employed and reading this with a small, smug sense of relief that none of the earlier sections apply to you, I’d gently ask you to check whether that relief is actually confidence, or just a slightly better-paid version of the same fear everyone else in this piece is dealing with.

“Just reskill” is the let-them-eat-cake of 2026

Every politician, every LinkedIn post, every corporate HR memo currently has the exact same one-word solution to all of this: reskill. Learn to code. Learn AI. Learn prompting. Pivot. Upskill. The word changes, the advice doesn’t. And I need to tell you, as gently as I can, that the actual research on whether this works is genuinely brutal.

Start with the gold standard. The US ran a properly randomised, controlled experiment on worker retraining between 1987 and 1992, called the National JTPA Study. It found that participants saw no statistically significant improvement in employment rates, earnings, or job stability compared to people who got no retraining at all, and whatever small gains did show up tended not to last. Brookings has a full breakdown of this and other retraining research. That’s not a hostile blogger’s opinion. That’s the actual controlled outcome of the US government’s own flagship retraining programme.

Move forward to the current version of that same programme, called WIOA, and a study covering over 23 million participation records between 2017 and 2023 found much the same thing. Only about 5% of eligible people even receive retraining under it. Of those who do, 45% end up going right back into the same industry they were retrained out of, and 27% end up in the exact same occupation. Whatever wage gains do show up look less like people acquiring genuinely new, AI-resistant skills, and more like a natural “catch-up” effect, where the most disadvantaged participants would have improved somewhat anyway, just from time passing.

And it’s not only a public-sector failure. Salesforce, one of the loudest corporate voices telling laid-off workers to go learn AI skills, offers its own affected employees free access to its own learning platform, Trailhead. Salesforce’s internal data, presented at its own 2025 conference, showed that fewer than 12% of employees who start a Trailhead learning path actually finish it, and of that small group who do finish, fewer than half ever end up using the new skill in their actual job. The full meta-analysis, including the Trailhead numbers, is here. Read that sequence again slowly: the company lays you off, hands you a free course as its version of an apology, and privately already knows that fewer than one in twenty people who take that course will ever use what it taught them. Everyone involved gets to call this “workforce investment” in a press release. Nobody has to call it what it actually is, which is a formality.

There’s an even crueller twist buried in a Harvard and NBER working paper: workers who specifically retrain into AI-exposed fields, the fields everyone is telling them to pivot toward because “that’s where the growth is,” end up earning roughly 29% less than comparable workers who don’t make that switch. Chasing the hot new skill often means arriving in a market that’s already crowded with people who had the exact same idea six months before you, or arriving into a field that gets automated again before you’ve even finished paying off the course.

I’m not telling you all this to make reskilling feel pointless, because it isn’t, and I’ll get to what actually differentiates the programmes that work in a minute. I’m telling you this because “just reskill” gets said with the confidence of a proven solution, when the actual evidence says it’s closer to a coin flip that’s usually rigged against you, unless it’s built a specific way. There’s one meaningful bright spot in the data worth mentioning here, so this section doesn’t read as pure nihilism: a nonprofit programme called Generation, which partners directly with employers who commit to actually hiring graduates before the training even begins, has produced over 16,000 graduates with an 82% job placement rate and 72% one-year retention. McKinsey’s writeup on Generation’s model is worth reading closely. The difference between that outcome and Trailhead’s 12% completion rate isn’t the quality of the content. It’s that Generation builds the employer relationship and the job at the end of the training into the programme itself, instead of handing you a course and hoping the market sorts the rest out. Remember that distinction. It matters enormously for the solutions section coming up.

Money alone won’t save you either

The other proposed fix you’ll hear a lot about, especially from the tech billionaires whose companies are doing a fair share of the displacing, is universal basic income. Just give everyone a cheque, the theory goes, and the whole problem dissolves.

Finland actually tested this properly, in a genuine randomised trial, giving 2,000 unemployed people 560 euros a month, unconditionally, for two years. The World Economic Forum’s summary of the results is a good, balanced starting point. The results, when they came in, told an extremely specific and important story. Recipients reported significantly better mental health, lower stress, and improved general wellbeing compared to the control group on standard unemployment benefits. That part is genuinely good news, and it’s not nothing. But on the metric the Finnish government actually cared most about, whether the payments got people back into meaningful employment, the effect was minimal to nonexistent. Stockton’s separate American pilot programme found something similar: modest, positive employment effects, alongside clearly documented improvements in wellbeing, but nothing resembling a wholesale solution to the identity vacuum we’ve spent this entire piece describing. Universal Basic Income’s global overview covers Stockton, Kenya, and several other pilots side by side, if you want to go deeper.

Put Finland and the madogiwazoku next to each other and a genuinely important pattern falls out. Money removes financial panic, and that matters enormously, don’t get me wrong, a person who isn’t terrified about rent is a person with far more room to think clearly. But money on its own does not replace Jahoda’s other four latent functions. It doesn’t hand you a time structure. It doesn’t hand you social contact. It doesn’t hand you a goal bigger than yourself. It doesn’t hand you a status that survives the question “so what do you do.” A basic income cheque and a madogiwazoku salary are, structurally, almost the exact same intervention: guaranteed money, with the actual human function of work left completely unaddressed. Both improve stress. Neither, on their own, rebuilds a self.

Which finally brings us to the only honest place this piece could have been heading the entire time.

Okay. Here’s what actually works.

Everything above was diagnosis. I said I’d get to treatment, so here it is, in full, without shortcuts. None of it is a hack, none of it is a course you can finish in a weekend, and I’m not going to pretend any of it makes this transition painless. But it’s built out of the actual gaps the research above exposed, not out of vibes, and not out of a course I’m secretly trying to sell you.

1. Separate your identity from your job title, starting today, not after the email arrives

This is the single highest-leverage thing on this entire list, and it’s also the one everyone skips because it doesn’t feel urgent until it’s too late to do calmly.

Right now, today, while you still have your job, sit down and actually answer the question “what do you do” without using your job title. Not your hobby version of the answer, the real one. If you genuinely cannot answer that question, that’s not a character flaw, it’s completely normal, and it’s also exactly the vulnerability the madogiwazoku research and the grief-and-identity research both point straight at. The goal isn’t to love your job less or work with less commitment. It’s to make sure your sense of self has at least one load-bearing wall that isn’t your employer’s org chart. People who already have that second wall in place, a serious hobby, a craft, a community role, a second identity of any real depth, consistently report an easier landing after job loss than people whose entire self-concept was sitting inside one company’s HR system.

2. Build the floor before the ceiling shows any sign of falling

I’ve written before, at length, about why a financial buffer isn’t an emergency fund so much as it’s what I call a “no fund,” the thing that buys you the right to say no to a bad decision made in panic. That logic matters here more than almost anywhere else. Every piece of research on retraining above found the same underlying pattern: it works far better when the person doing it isn’t simultaneously terrified about next month’s rent. Panic narrows your options and pushes you toward the fastest bad job instead of the right slower one. A buffer, even a modest one, is what lets you actually use a genuinely good retraining programme like Generation instead of grabbing the first gig site that’ll have you, out of sheer fear.

3. Stop competing with AI and start operating it

The data above is fairly consistent on this point across every single industry it touched, from tech to law to journalism to healthcare: the roles disappearing fastest are the purely repetitive, rules-based, low-judgement ones, the exact tasks a model can be trained on from twenty years of transcripts. The roles growing fastest are the ones that pair a technical skill with judgement, taste, accountability, and the willingness to be the human who signs off on the machine’s output. Practically, that means the safest bet in an AI-saturated field isn’t refusing to touch the tools out of principle, the way a modern-day Luddite might smash the loom instead of learning to run it. It’s becoming the person who directs the tool, checks it, and catches the specific, expensive mistakes it still reliably makes. The junior coder who tests AI-generated code and traces its failures is in a materially different position than the junior coder trying to out-type the AI at writing that same code from scratch. The paralegal who becomes the person who catches the AI’s confidently wrong citation is worth more than the paralegal who refuses to open the tool at all.

4. Deliberately rebuild the five things your job was quietly giving you

Go back to Jahoda’s list, because it doubles as the single best personal checklist for surviving unemployment I’ve ever come across, and almost nobody uses it that way. Time structure, social contact, a purpose beyond yourself, a status and identity, and regular activity. If a job disappears, don’t wait for a new one to hand these back to you automatically. Build them on purpose, in parallel with the job search, not after it. A fixed daily schedule you keep whether or not anyone’s watching. A standing commitment with other people, ideally people outside your former industry, so your entire social world isn’t grieving the same loss at the same time you are. A project, even a small, unpaid one, that gives your days a “why” beyond “finding the next pay cheque.” Isolation is the single most consistently documented factor that makes unemployment’s psychological toll worse, and it’s also the factor almost entirely within your own control to fight, independent of whether the job market cooperates.

5. If you’re choosing where to retrain, choose the programme with a job already attached to the end of it

Given everything the Trailhead and WIOA numbers told us, the single most important filter for any reskilling decision isn’t “is this the trendiest skill.” It’s “does this specific programme have an actual employer relationship, or a track record of actual placements, sitting at the end of it.” Generation’s 82% placement rate versus Trailhead’s 12% completion rate isn’t a coincidence of content quality. It’s the difference between a programme designed around getting you hired and a course designed around making a company’s layoffs look responsible in a press release. Ask that one blunt question before you spend a single evening on any course: who, specifically, is going to hire the person who finishes this, and how do you know.

6. Widen the definition of “your industry” before you’re forced to

The hourglass shape we talked about earlier, thick at the senior end, thick at the very bottom of cheap labour, thin in the AI-automatable middle, isn’t unique to Indian IT. It’s showing up across law, journalism, tech, and administrative work globally. Which means the honest long-term move for a lot of people currently in the squeezed middle isn’t clinging tighter to the exact job description on their old business card. It’s looking sideways, toward roles that pair your existing domain knowledge with the judgement AI still can’t reliably fake: healthcare, skilled trades, anything requiring physical presence and real human trust, or roles inside your own current industry that involve supervising and correcting AI output rather than competing with it directly. None of that is instant. Almost none of it is comfortable. But it’s a materially better bet than hoping the specific job you have today survives completely unchanged.

7. Diversify your income the way you’d diversify a portfolio, before you need to, not after

One quiet thread running through almost every survival story I found while researching this piece is that the people who landed softest weren’t necessarily the most talented, they were the ones who already had a second, smaller income stream running before the layoff hit. Freelance work on the side. A small teaching gig. A newsletter, a course, a craft sold on weekends. None of these need to replace your main income while you have one. Their entire value is that they exist at all, because the day the main income disappears, you’re not starting completely from zero, you’re scaling something that already has a pulse. This is boring advice. It is also, based on everything in the retraining and UBI data above, some of the only advice in this entire piece with a genuinely strong evidence base behind it.

8. If you’re the one who stayed, don’t waste the survivor’s guilt, use it

If you read the section on layoff survivor syndrome and recognised yourself in it, here’s the reframe. That low hum of “I could be next” is unpleasant, but it’s also accurate information, and accurate information used early is worth infinitely more than the same information received later, over email, with no notice. Don’t spend the anxiety doing nothing. Spend it doing steps one through seven while you still have a salary funding the runway. The single worst time to build a financial buffer, a second skill, or a second identity is after you need one. The best time was months ago. The second best time is this week.

To whoever’s refreshing their layoff Slack channel right now

I want to end this the same honest way I started it, because I think you’ve earned that by making it this far.

I don’t know if your specific job survives the next five years. Genuinely, nobody does, not the WEF, not Goldman Sachs, not the confident man on LinkedIn selling a course about it. The data in this piece is real, and it’s also, by its nature, already slightly out of date the moment I hit publish, because that’s how fast this particular transition is moving.

What I do know, because the research on it is unambiguous even where the job-market research isn’t, is this: the loss of a job is survivable. Humans have survived it for as long as jobs have existed, from the weavers of Nottingham to the tellers who watched the ATM arrive to Ravi, staring at an email in Bengaluru. What actually determines whether you come out the other side intact isn’t the size of your severance cheque or how quickly you land the next role. It’s whether you built something, anything, structure, purpose, people, identity, that didn’t live entirely inside your employer’s server. The horse had no say in any of this. You do. Not a full say, not a comfortable one, but more than zero, and more than the madogiwazoku sitting quietly by his window, fully paid and fully forgotten.

So build the second wall now, while you still have the calm to build it properly, rather than in a panic after the email arrives. Not because the email is definitely coming. Because if it does, you deserve to meet it as someone who was never just a job title to begin with.

Further Reading and Sources

Discover more from niranjanpathak.info

Subscribe to get the latest posts sent to your email.