Essay · AI & Work
AI is a productivity enhancer, not a replacement. The headlines say otherwise. Four decades across eight industries say the headlines are wrong · Juan Vegarra · December 2026
SHARED ON LINKEDIN
Every morning the same headline runs with a different logo: record revenue, thousands of layoffs, AI named as the reason.
My career started in 1986 writing the software for a predictive scoring model: mathematical regression, the agent of its day, run against a large historical clinical dataset, the training corpus of its day. An early version of exactly what we now call AI, four decades before the name. So let me offer the view from someone who has actually replaced human tasks with software and lived with what came next.
The headline is right that work is changing. It is wrong about almost everything else.
The task is not the job. The model produced the score; it did not replace the doctor, because producing the score was never the doctor's job. Judgment was. AI is extraordinary at the task and helpless at the job.
We have seen this film before. The loom did not end textile work. The spreadsheet grew the number of accountants. The ATM grew the number of tellers for decades. Every industrial revolution destroys a category of work and creates more than it took, and every time we forget until the next one.
AI is a productivity enhancer, not a replacement. The company that builds the future is not the one with the fewest people. It is the one whose people, armed with the machine, can do what neither could alone.
Every morning this year the same headline runs with a different logo on it: a company reports record revenue, announces thousands of layoffs, and names artificial intelligence as the reason. The tracker I looked at this week counted more than two hundred thousand roles cut in 2026, with AI cited in over half of them. If you read only the headlines, the conclusion is obvious and terrifying: the machines have arrived, and they are here for your desk.
I have watched a version of this movie four times, in four industries, over a career that started in 1986 writing a predictive scoring model on a hospital research team: software that ran mathematical regression against a large historical clinical dataset to predict how a patient would fare. An agent reasoning over data at scale to produce a prediction, the same shape as what people now call AI, decades before the name existed. So let me offer the view from someone who has actually replaced human tasks with software and lived with the consequences. The headline is not wrong that jobs are changing. It is wrong about almost everything else, and the gap between those two things is where your future actually lives.
Start with the layoffs themselves, because even they do not say what the headline claims. Deutsche Bank analysts have a name for the pattern this year: AI redundancy washing. Companies that overhired during the pandemic, or that face investor pressure to cut costs, announce the layoffs they were going to make anyway and dress them in the language of AI, because AI is the fashionable reason and it moves the stock. Even the CEO of the most famous AI company has said out loud that some firms are blaming AI for cuts that were really about cost. When Microsoft cut roles this year it stated plainly that the jobs were not being replaced by AI; the work was simply changing. The most honest read of the data is not machines replacing people. It is a smaller round of cost-cutting wearing a costume the market rewards.
And where AI genuinely is doing the work, look at what happened when companies tried the pure version. The firms that made headlines replacing customer service entirely with AI quietly watched satisfaction scores collapse on the calls that actually matter, the fraud dispute, the billing crisis, the frightened customer, and walked much of it back. Replacement is easy to announce and hard to live with. The task the demo handled was never the whole job.
This is the distinction the entire panic misses, and it is the one I learned in 1986. I wrote the software for a predictive scoring model: mathematical regression run against a large historical dataset to estimate how a patient would fare. The regression was the agent of its day; the historical dataset was its training corpus. The model did real intellectual work, the kind that used to take a trained human, and it did it faster and at a scale no person could match. And still it did not replace the doctor, because producing the score was never the doctor's job. Judgment was. The machine did the predicting so the physician could do the deciding, and the deciding is where the value always lived.
Every serious study lands in the same place once you get past the headline. The consensus is not wholesale replacement; it is that most exposed roles will be augmented, not eliminated, and that the roles where AI assists expert humans are growing faster and paying better than the ones it does not touch. AI is extraordinary at the task and helpless at the job, because a job is a bundle of tasks held together by judgment, relationship, accountability, and the thousand small acts of physical and human intervention that no model performs. The nurse who reads the room. The electrician in the crawlspace. The banker the credit model cannot replace because someone has to be accountable when the model is wrong. Whole fields run on manual intervention that no amount of intelligence, artificial or otherwise, removes.
None of this is unprecedented. It is the oldest pattern in industrial history, and we have simply forgotten it because the last full turn happened before most of us were working. The mechanical loom did not end textile work; it ended one kind of it and created factory work, then design work, then whole industries the weavers could not have imagined. The spreadsheet was supposed to end accounting. Instead the number of accountants grew, because when analysis got cheap, businesses wanted vastly more of it, and the accountant moved up the stack from arithmetic to advice. The automated teller machine was supposed to end the bank teller. The number of tellers rose for decades afterward, because cheaper branches meant more branches, and the teller's job shifted from counting cash to serving people.
The pattern is always the same, and it is always misread in the moment. A technology destroys a category of task, the jobs built narrowly on that task suffer, and meanwhile the productivity it unleashes creates demand, and new work, that the people panicking could not have named in advance. This is the paradigm shift that has run continuously since the Industrial Revolution, and AI is its newest and most dramatic chapter, not an exception to it. The losses are real and they land on real people; I am not waving those away. But the ledger has two columns, and the headline only ever prints one.
If the task is not the job and the pattern is not new, then the move is not to fear the tool; it is to become the person who wields it. The workers who are hardest to replace, the research keeps finding, are the ones who pair domain judgment with AI fluency, and multiply their output well beyond peers who bring neither. That is not a threat. That is the best productivity offer of your career, if you take it. Learn the tool the way your parents learned the computer: not to compete with it, but to stop doing by hand the parts of your work that were never the point, so you can do more of the part that was.
The real risk in this moment is not that AI takes every job. It is that we believe the headline, treat a productivity revolution as a replacement event, and cut the humans whose judgment we will need precisely when the tool starts making confident mistakes. The company building the future is not the one with the fewest people. It is the one whose people, armed with the machine, can do what neither could alone.
AI does the task. A human still holds the job, because judgment, accountability, and the hand that reaches into the physical world were never the parts a machine could take. Every industrial revolution destroys work and creates more of it. This one is no different, except in speed, and speed rewards the person who learns the tool first.
Juan Vegarra is the author of An Outsider's Playbook (forthcoming). More from the Notebook · Continue the conversation on LinkedIn
READ NEXT