{"course":{"slug":"ai-actually-explained","title":"AI, Actually Explained","url":"https://addaly.com/learn/ai-actually-explained","level":"beginner","track":"start"},"lesson":{"slug":"the-training-ladder","title":"The rung that goes missing","url":"https://addaly.com/learn/ai-actually-explained/the-training-ladder","position":77,"of":92,"minutes":10,"answer":"Professions have always made their seniors by giving juniors the routine tasks a model now does.","keyIdea":"Professions have always made their seniors by giving juniors the routine tasks a model now does, and payroll data since 2022 shows employment for the youngest workers in the most exposed occupations falling while their older colleagues' holds, and nobody has yet shown how the next seniors get made."},"body":{"format":"markdown","text":"## How a senior was made\n\nThink about how anyone became good at a knowledge job. A trainee lawyer spent two years reviewing documents and drafting the standard clauses. A junior developer fixed small bugs and wrote the boilerplate. A junior journalist rewrote press releases. A trainee translator did the first pass and a senior corrected it. A junior analyst built the spreadsheet and a senior read it. The routine task was not just the junior's output; it was the junior's education. Doing the small thing badly and being corrected is how the judgement to do the big thing was formed.\n\nNow look at the exposure lesson's list of what a model does well: drafting, summarising, first-pass translation, boilerplate code, standard clauses, document review. It is the same list. The tasks that train a junior are the tasks a model does first.\n\n> Figure: How a profession used to make its seniors. Steps: 1. The routine task (Document review, standard clauses, boilerplate code, the first-pass translation, the basic spreadsheet); 2. Done badly by a junior (Slowly, imperfectly, at a cost the firm absorbed as training); 3. Corrected by a senior (The correction, not the task, is where the judgement was transferred); 4. Repeated for two years (Until the junior could tell a good answer from a plausible one without being told); 5. A senior exists (Able to check work, carry the consequence, and correct the next junior. Nobody has shown another route to this). The routine task was not only the junior's output; it was the junior's education. It is also the first thing a model does well, so the rung that made the next generation is the rung under pressure. The ladder still stands for everyone already on it and becomes hard to climb from the ground.\n\n## What the data began to show\n\nFor two years this was an argument about what might happen. In 2025 a group at Stanford looked at payroll records covering millions of American workers, month by month, from before the chatbots to the present.\n\nThey found that employment for workers aged twenty-two to twenty-five in the occupations most exposed to AI (software development and customer service among them) had fallen by around thirteen per cent relative to less-exposed occupations since late 2022. Workers over thirty in the same occupations held steady or grew. In occupations where the tools mostly complement a worker rather than replace their tasks, the young did fine. The decline was specifically at the entry rung, specifically where the tool substitutes.\n\nThe study is one study, in one country, with the usual limits: the same years saw interest rates rise, a correction to pandemic over-hiring, and a general cooling in software employment that has causes beyond AI. The authors controlled for what they could and were careful in their claims. But the shape they found is exactly the shape the ladder argument predicts, and it appeared in the data before most firms had said out loud what they were doing.\n\n## What firms are doing\n\nThree things, and it is worth naming them without judgement because each is a rational response to the same pressure.\n\nSome have reduced graduate intake. If the junior's output is now the model's, and the junior costs a salary, the arithmetic is obvious in the quarter it is done. Announcements from large firms in 2023 and 2024 of a pause in hiring for roles the tools could cover were of this kind.\n\nSome have kept intake and changed the job. The junior supervises the model's draft rather than producing one, learns to check rather than to write, and is pushed toward judgement earlier. Whether checking without ever having produced actually builds judgement is the open question, and the honest answer is that nobody knows yet.\n\nSome have kept intake deliberately, on the argument that the seniors of 2032 have to come from somewhere and that a firm which stops making them will be buying them at a premium later. This view is common in professions with long training pipelines and rare in those without.\n\n## Why this is the thing to worry about\n\nNot mass unemployment. The exposure studies, the experiments and the payroll data agree that the effect on most existing workers is a change in tasks, not a loss of employment. The specific thing that is breaking is the mechanism by which a trade renews itself. If the rung is gone, the ladder still stands for everyone already on it and becomes impossible to climb from the ground. That is a slower harm than a layoff and a larger one, and it does not show up in this year's figures. It shows up in the age distribution of a profession ten years on.\n\n## What is unknown\n\nWhether juniors can learn judgement by supervising rather than doing: the maths-tutoring experiment from an earlier module suggests they cannot, if the tool does the work, and can, if it only helps. Whether the entry-level decline is AI or the economic cycle; the Stanford data says AI, and it is one dataset. Whether firms will act on a ten-year problem in a quarterly system; history is not encouraging. And whether new entry roles appear that train people in the new shape of the work, as the spreadsheet created analysts. The next lesson is about what earlier automations actually did, because that record is the best evidence about the part of this nobody has lived through yet."},"check":{"question":"Payroll data since 2022 shows employment for 22-to-25-year-olds falling in the most AI-exposed occupations while older workers in the same occupations hold steady. Why would the effect land on the youngest?","options":[{"text":"Younger workers are less skilled with the tools, so firms replace them first.","correct":false,"tempting":"Inverts the evidence; the young are the heaviest users, and the experiments show novices gain most from the tools."},{"text":"Entry-level roles consist of the routine tasks the model does first, so their output is covered and the senior's is not.","correct":true},{"text":"Older workers have contracts that make them harder to dismiss.","correct":false,"tempting":"Reaches for a legal cause; the pattern held across occupations with very different employment terms."},{"text":"Firms are hiring fewer people of every age, and the young simply show it first because they are the ones who would have been hired.","correct":false,"tempting":"Contradicts the data; in the same period and occupations, older workers' employment held or grew."}]},"prev":{"slug":"who-gains-first","title":"Novices gain first, and what that means for everyone else","url":"https://addaly.com/learn/ai-actually-explained/who-gains-first"},"next":{"slug":"what-the-atm-did","title":"What the cash machine did to bank tellers","url":"https://addaly.com/learn/ai-actually-explained/what-the-atm-did"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'The rung that goes missing', in AI, Actually Explained. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-actually-explained/the-training-ladder (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"The rung that goes missing","courseTitle":"AI, Actually Explained","module":{"index":8,"slug":"ai-and-jobs","title":"AI and work, with the evidence"},"answer":null,"keyIdea":"Professions have always made their seniors by giving juniors the routine tasks a model now does, and payroll data since 2022 shows employment for the youngest workers in the most exposed occupations falling while their older colleagues’ holds, and nobody has yet shown how the next seniors get made.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"flow","steps":[{"label":"The routine task","detail":"Document review, standard clauses, boilerplate code, the first-pass translation, the basic spreadsheet"},{"label":"Done badly by a junior","detail":"Slowly, imperfectly, at a cost the firm absorbed as training"},{"label":"Corrected by a senior","detail":"The correction, not the task, is where the judgement was transferred"},{"label":"Repeated for two years","detail":"Until the junior could tell a good answer from a plausible one without being told"},{"label":"A senior exists","detail":"Able to check work, carry the consequence, and correct the next junior. Nobody has shown another route to this"}],"title":"How a profession used to make its seniors","caption":"The routine task was not only the junior’s output; it was the junior’s education. It is also the first thing a model does well, so the rung that made the next generation is the rung under pressure. The ladder still stands for everyone already on it and becomes hard to climb from the ground."},"alt":"Figure: How a profession used to make its seniors. Steps: 1. The routine task (Document review, standard clauses, boilerplate code, the first-pass translation, the basic spreadsheet); 2. Done badly by a junior (Slowly, imperfectly, at a cost the firm absorbed as training); 3. Corrected by a senior (The correction, not the task, is where the judgement was transferred); 4. Repeated for two years (Until the junior could tell a good answer from a plausible one without being told); 5. A senior exists (Able to check work, carry the consequence, and correct the next junior. Nobody has shown another route to this). The routine task was not only the junior’s output; it was the junior’s education. It is also the first thing a model does well, so the rung that made the next generation is the rung under pressure. The ladder still stands for everyone already on it and becomes hard to climb from the ground."}},"blocks":[{"t":"h2","id":"how-a-senior-was-made","runs":[{"text":"How a senior was made"}]},{"t":"p","runs":[{"text":"Think about how anyone became good at a knowledge job. A trainee lawyer spent two years reviewing documents and drafting the standard clauses. A junior developer fixed small bugs and wrote the boilerplate. A junior journalist rewrote press releases. A trainee translator did the first pass and a senior corrected it. A junior analyst built the spreadsheet and a senior read it. The routine task was not just the junior’s output; it was the junior’s education. Doing the small thing badly and being corrected is how the judgement to do the big thing was formed."}]},{"t":"p","runs":[{"text":"Now look at the exposure lesson’s list of what a model does well: drafting, summarising, first-pass translation, boilerplate code, standard clauses, document review. It is the same list. The tasks that train a junior are the tasks a model does first."}]},{"t":"h2","id":"what-the-data-began-to-show","runs":[{"text":"What the data began to show"}]},{"t":"p","runs":[{"text":"For two years this was an argument about what might happen. In 2025 a group at Stanford looked at payroll records covering millions of American workers, month by month, from before the chatbots to the present."}]},{"t":"p","runs":[{"text":"They found that employment for workers aged twenty-two to twenty-five in the occupations most exposed to AI (software development and customer service among them) had fallen by around thirteen per cent relative to less-exposed occupations since late 2022. Workers over thirty in the same occupations held steady or grew. In occupations where the tools mostly complement a worker rather than replace their tasks, the young did fine. The decline was specifically at the entry rung, specifically where the tool substitutes."}]},{"t":"p","runs":[{"text":"The study is one study, in one country, with the usual limits: the same years saw interest rates rise, a correction to pandemic over-hiring, and a general cooling in software employment that has causes beyond AI. The authors controlled for what they could and were careful in their claims. But the shape they found is exactly the shape the ladder argument predicts, and it appeared in the data before most firms had said out loud what they were doing."}]},{"t":"h2","id":"what-firms-are-doing","runs":[{"text":"What firms are doing"}]},{"t":"p","runs":[{"text":"Three things, and it is worth naming them without judgement because each is a rational response to the same pressure."}]},{"t":"p","runs":[{"text":"Some have reduced graduate intake. If the junior’s output is now the model’s, and the junior costs a salary, the arithmetic is obvious in the quarter it is done. Announcements from large firms in 2023 and 2024 of a pause in hiring for roles the tools could cover were of this kind."}]},{"t":"p","runs":[{"text":"Some have kept intake and changed the job. The junior supervises the model’s draft rather than producing one, learns to check rather than to write, and is pushed toward judgement earlier. Whether checking without ever having produced actually builds judgement is the open question, and the honest answer is that nobody knows yet."}]},{"t":"p","runs":[{"text":"Some have kept intake deliberately, on the argument that the seniors of 2032 have to come from somewhere and that a firm which stops making them will be buying them at a premium later. This view is common in professions with long training pipelines and rare in those without."}]},{"t":"h2","id":"why-this-is-the-thing-to-worry-about","runs":[{"text":"Why this is the thing to worry about"}]},{"t":"p","runs":[{"text":"Not mass unemployment. The exposure studies, the experiments and the payroll data agree that the effect on most existing workers is a change in tasks, not a loss of employment. The specific thing that is breaking is the mechanism by which a trade renews itself. If the rung is gone, the ladder still stands for everyone already on it and becomes impossible to climb from the ground. That is a slower harm than a layoff and a larger one, and it does not show up in this year’s figures. It shows up in the age distribution of a profession ten years on."}]},{"t":"h2","id":"what-is-unknown","runs":[{"text":"What is unknown"}]},{"t":"p","runs":[{"text":"Whether juniors can learn judgement by supervising rather than doing: the maths-tutoring experiment from an earlier module suggests they cannot, if the tool does the work, and can, if it only helps. Whether the entry-level decline is AI or the economic cycle; the Stanford data says AI, and it is one dataset. Whether firms will act on a ten-year problem in a quarterly system; history is not encouraging. And whether new entry roles appear that train people in the new shape of the work, as the spreadsheet created analysts. The next lesson is about what earlier automations actually did, because that record is the best evidence about the part of this nobody has lived through yet."}]}],"check":{"question":[{"text":"Payroll data since 2022 shows employment for 22-to-25-year-olds falling in the most AI-exposed occupations while older workers in the same occupations hold steady. Why would the effect land on the youngest?"}],"options":[{"text":[{"text":"Younger workers are less skilled with the tools, so firms replace them first."}],"correct":false,"tempting":[{"text":"Inverts the evidence; the young are the heaviest users, and the experiments show novices gain most from the tools."}]},{"text":[{"text":"Entry-level roles consist of the routine tasks the model does first, so their output is covered and the senior’s is not."}],"correct":true,"tempting":null},{"text":[{"text":"Older workers have contracts that make them harder to dismiss."}],"correct":false,"tempting":[{"text":"Reaches for a legal cause; the pattern held across occupations with very different employment terms."}]},{"text":[{"text":"Firms are hiring fewer people of every age, and the young simply show it first because they are the ones who would have been hired."}],"correct":false,"tempting":[{"text":"Contradicts the data; in the same period and occupations, older workers’ employment held or grew."}]}],"order":[2,3,1,0],"why":null},"prev":{"slug":"who-gains-first","title":"Novices gain first, and what that means for everyone else","minutes":9},"next":{"slug":"what-the-atm-did","title":"What the cash machine did to bank tellers","minutes":9}}}