Is AI going to take my job?

The full story

The episode's narration, word for word. Numbers link to the sources below; timestamps play that moment.

The question 0:00

Last season ended on a question you may have asked yourself. The headlines swing from mass unemployment to nothing to see here. This season, AI and Us, we go to the evidence instead, and we say what each study actually measured. So let's start with the big one. Is AI going to take my job?

Forty per cent, or four? 0:23

Start with two numbers. In January 2024, the head of the International Monetary Fund, Kristalina Georgieva, wrote that almost forty per cent of jobs worldwide are exposed to AI ¹. In August 2025, the government agency Jobs and Skills Australia found that only about four per cent of Australian workers are in jobs highly exposed to automation by generative AI ². Both are careful studies, but they count different things: the IMF, jobs it ranked as highly exposed to AI, whether AI might help or replace people; the Australian agency, only jobs scoring high for automation ^3,2.

What exposed means 1:08

To measure exposure, researchers break each job into tasks, and score each task for how much of it generative AI could do ^4,2. The International Labour Organization's 2025 index surveyed over sixteen hundred workers in Poland, checked with experts, then built an AI assistant to score the rest ⁴. Notice the word could. Exposure isn't a forecast that employers will use it, or a measure of how well AI would do the task ^4,2.

A job is a bundle 1:40

Take a made-up receptionist's week. Phones, bookings, emails, filing, and greeting people at the desk. Give each task a score, weight it by its hours, and this job comes out about half exposed, by our count. But look at which half. Bookings and emails score high, while greeting a person barely scores at all. That's why the ILO expects most jobs to change shape rather than vanish: most are made of tasks that still need a person ⁴.

Automation or augmentation 2:13

A high score can mean two different things. Automation means the same work gets done with fewer human hours ². Augmentation means a person gets more, or better, work done in the same time ². And if people then want more of that work, the hours may not fall at all ². Jobs and Skills Australia found nearly half of workers are in jobs with low automation but medium augmentation, which it reads as change, rather than disruption ².

Who is most exposed? 2:44

So who's most exposed? In both the global and the Australian studies, it's clerical and administrative work, much of which escaped earlier waves of automation ^4,2. Exposure falls more on women. In high-income countries, the ILO puts nine point six per cent of women's jobs in its highest exposure group, against three point five per cent of men's ⁴. And richer countries are more exposed overall: about a third of jobs, against about a tenth in low-income countries ⁴.

And least exposed? 3:17

And the least exposed? Jobs and Skills Australia modelled the economy out to 2050 ². In its scenario, compared with a future without generative AI, the jobs that gain most include cleaners, nurses and midwives, construction labourers and hospitality workers ². Clerks, receptionists and bookkeepers lose most, though total employment ends up higher ². The agency calls these illustrative scenarios, built on assumptions, so they aren't predictions ².

The cash machine 3:50

Have we been here before? In the 1970s, banks in the United States began installing automated teller machines, which handed out cash and took deposits ⁵. From the mid-1990s they spread fast, and by 2015 there were over four hundred thousand in the US alone, the economist James Bessen wrote ⁵. You'd expect teller jobs to collapse. But as the machines spread, the number of teller jobs did not fall ⁵.

Why the tellers stayed 4:20

Bessen gives two reasons ⁵. First, cash machines made branches cheaper to run ⁵. In the average city market, a branch needed thirteen tellers in 2004, down from twenty in 1988, but banks responded by opening more: urban branches rose forty-three per cent ⁵. Second, the job changed: handling cash mattered less, and looking after customers mattered more ⁵.

Where the analogy breaks 4:50

But don't lean on this story too hard. Now online banking is cutting branch numbers, and the US Bureau of Labor Statistics projects teller jobs will fall thirteen per cent between 2025 and 2035 ⁶. And a cash machine did a few tasks in one job, while generative AI's exposure touches one in four jobs worldwide ^4,5. Whether work grows or shrinks depends partly on whether people want more of what gets cheaper ^2,5.

A call centre, 2020 5:21

So what has actually been measured? Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed more than five thousand customer support agents at a Fortune 500 software company, mostly working from the Philippines ⁷. From 2020, they were given an assistant built on a GPT language model, which suggested replies they could use, edit or ignore ⁷. On average, issues resolved per hour rose fifteen per cent ⁷.

Who it helped 5:53

But the average hides the story. The less experienced and lower-skilled agents improved by about thirty per cent ⁷. New agents with the assistant were as good after two months as agents without it after six ⁷. The most experienced barely changed, and the quality of their chats dipped slightly ⁷. Note what this measured, though: output per hour in one company, not whether anyone lost a job ⁷.

The canaries 6:20

For jobs, you need employment records. Brynjolfsson, Bharat Chandar and Ruyu Chen used US data covering millions of workers up to June 2026 ⁸. They found no sign of economy-wide job losses ⁸. But for twenty-two to twenty-five year olds in the more AI-exposed jobs, like software development and customer service, employment fell about eleven per cent from late 2022, while in less exposed jobs it grew about ten ⁸. That leaves them nineteen per cent behind, mostly through less hiring ⁸.

What the canaries can't tell us 6:59

The authors call these early warning signs, not proof that AI caused them ⁸. Some of the gap began before ChatGPT, and it shrinks once you account for education ⁸. Their exposure scores come from a study co-written by OpenAI researchers, and from usage data published by Anthropic, which makes Claude ^8,9. Declines were concentrated where people use AI to do tasks for them, not where it assists ⁸.

Elsewhere, not yet 7:28

Other studies find less ^10,11. In Denmark, the economists Anders Humlum and Emilie Vestergaard linked surveys of chatbot use to official earnings records ¹¹. Two years after ChatGPT's launch, they found no effect on earnings or hours, ruling out anything over two per cent; what changed was how work was organised ¹¹. And on the fourteenth of October 2026, Yale's Budget Lab reported that the US job mix still isn't shifting in ways that clearly line up with AI ¹⁰.

And in Australia 8:03

In Australia, Jobs and Skills Australia found in August 2025 only limited job displacement so far ². A few narrow occupations, such as voice-over artists, have seen less demand, but most change was roles evolving and people retraining ². Unlike the US payroll study, it found entry-level hiring still active in job ads up to 2025 ². It adds that entry-level jobs could still be affected later ².

Big forecasts, small ones 8:35

Forecasts disagree even more. In April 2023, the Goldman Sachs economists Joseph Briggs and Devesh Kodnani wrote that generative AI could expose the equivalent of three hundred million full-time jobs to automation, and lift global GDP by seven per cent in ten years ¹². Even they expected most jobs to be complemented rather than replaced ¹². In 2024, the economist Daron Acemoglu estimated about one per cent for the US ¹³.

Acemoglu's arithmetic 9:11

His reasoning is simple arithmetic. He takes about a fifth of US work tasks as exposed to AI, and, borrowing an estimate made for computer vision, takes only about twenty-three per cent of those as profitable to hand over within ten years ¹³. That's under five per cent of all tasks, by our count ¹³. Multiply by an average cost saving of about fourteen per cent, and productivity rises at most about two-thirds of one per cent ¹³. Change those assumptions and the answer changes: that's the real argument.

The bosses disagree too 9:47

In May 2025, Anthropic's chief executive, Dario Amodei, told Axios that AI could wipe out half of all entry-level white-collar jobs, and push unemployment to ten to twenty per cent within five years ¹⁴. In June, Nvidia's chief executive, Jensen Huang, said he pretty much disagreed with almost everything Amodei says: some jobs will become obsolete, but many will be created ¹⁵. Both sell AI or its chips, and both were forecasting ^14,15. And this channel uses AI too. The stills in this video are made by an open image model called FLUX, on a laptop ¹⁶. The scripts are drafted with AI tools, including Claude, made by Anthropic, one of the companies in this story, and every claim is checked against the sources linked below ¹⁷.

So, is it coming for my job? 10:45

So, is AI going to take your job? As of October 2026, many jobs contain tasks AI could help with or do, but exposure isn't replacement ^1,4,2. Measured effects are small overall, with one warning sign: fewer young people hired into exposed jobs in the US ^8,10,11. What happens next depends on which tasks change, whether demand grows, and choices by employers and governments ^2,5. As Jobs and Skills Australia puts it, the outcomes are not predetermined ².

The season ahead 11:20

This season asks: How much energy and water does it use? Can you still tell what's real? Why is it sometimes unfair, can it be creative, can it do real science, does it understand anything? Who writes the rules, what worries its builders, how do you use it well, and will it ever be as smart as us? But first, these models were trained on other people's work. So who owns what AI learned from?

Sources

Every factual claim in the episode is tied to one of these. Spotted an error? Tell us.

1. Kristalina Georgieva, IMF Blog, AI Will Transform the Global Economy. Let's Make Sure… — imf.org
2. Jobs and Skills Australia, Our Gen AI Transition: Implications for Work and Skills (14… — jobsandskills.gov.au
3. Cazzaniga et al., Gen-AI: Artificial Intelligence and the Future of Work, IMF Staff… — imf.org
4. International Labour Organization, Generative AI and Jobs: A Refined Global Index of… — ilo.org
5. James Bessen, Toil and Technology, IMF Finance and Development (March 2015) — imf.org
6. US Bureau of Labor Statistics, Occupational Outlook Handbook: Tellers (accessed 16… — bls.gov
7. Erik Brynjolfsson, Danielle Li and Lindsey R. Raymond, Generative AI at Work (arXiv… — arxiv.org
8. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, Canaries in the Coal Mine? Six Facts… — digitaleconomy.stanford.edu
9. Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock, GPTs are GPTs: An Early… — arxiv.org
10. The Budget Lab at Yale, Tracking the Impact of AI on the Labor Market (updated 14… — budgetlab.yale.edu
11. Anders Humlum and Emilie Vestergaard, Still Waters, Rapid Currents: Early Labor Market… — nber.org
12. Goldman Sachs, Generative AI could raise global GDP by 7% (5 April 2023) — goldmansachs.com
13. Daron Acemoglu, The Simple Macroeconomics of AI, NBER Working Paper 32487 (May 2024) — nber.org
14. Axios, Behind the Curtain: A white-collar bloodbath (28 May 2025) — axios.com
15. Fortune, Nvidia's Jensen Huang says he disagrees with almost everything Anthropic CEO… — fortune.com
16. Black Forest Labs, FLUX.1 [schnell] model card — huggingface.co
17. Anthropic, Claude — anthropic.com

Researched and scripted with AI assistance, fact-checked claim by claim, with synthetic narration and diagrams drawn in code. How we make episodes.