← Research
Research · Essay

The Missing Rungs

AI is taking the entry-level work people used to climb through.

Last reviewed: 9 November 2026 · Next review due: 9 November 2027

The missing rungs are the lower rungs of the career ladder: the entry-level tasks people used to climb through on the way to senior judgement, automated away before anyone noticed they were load-bearing. Rahim Hirji named the pattern on 21 September 2025, in "The Missing Rungs: What Nobody Will Tell You About AI and Your Job": "we're not facing mass unemployment. We're facing what I'm calling Missing Rungs Problem", which he glosses in the next breath as the systematic removal of how people advance up the career ladder. The essay's own form of words is the Missing Rungs Problem; this research uses the shorter form. He develops it in SuperSkills (Kogan Page, 2026). Earlier private or spoken use cannot be excluded. See the glossary.

Questions this page answersQuestions this page partly answersAll 1735 questions this research covers

When AI takes over entry-level work, the first few rungs of the career ladder go with it. Those rungs were never only production. They were where people learned to be senior, by doing small things badly, being corrected, and doing them again. Take them out and an organisation gets juniors who look productive on every dashboard and a senior bench that is quietly not being built.

Definition

The missing rungs: the lower rungs of the career ladder, meaning the entry-level tasks people used to climb through on the way to becoming senior, automated away before anyone noticed they were load-bearing. Rahim Hirji's term. He names it in the first person on 21 September 2025, in "The Missing Rungs": "what I'm calling Missing Rungs Problem". Developed in SuperSkills (Kogan Page, 2026). Earlier private or spoken use cannot be excluded.

Share this definition as a card

The Missing Rungs Problem, defined#

> The Missing Rungs Problem. The systematic disappearance of entry-level and early-career work experiences through which employees historically built the judgement, skill and relationships that made them ready for senior roles. Caused by AI and automation absorbing the formative tasks that once served as the training ground for professional growth. > > Why it matters. Without these developmental experiences, organisations produce employees who can deliver AI-augmented output today but cannot independently lead, decide or problem-solve tomorrow. The result is a leadership pipeline that looks full on paper but is structurally empty. > > How it shows up. Declining internal promotion rates. Growing dependency on a small number of senior staff who hold irreplaceable institutional knowledge. Junior work that is done faster and learned from less. > > The opposite. Deliberately designed early-career pathways where AI augments learning rather than replacing it, and where organisations measure capability development alongside productivity. > > Rahim Hirji, first published 21 September 2025

What actually goes missing#

Professions without apprentices#

A law firm with plenty of seasoned partners but no associates being trained underneath them. A hospital with senior surgeons but a shortage of residents coming up the ranks. That is the end state this describes, and it arrives without anybody deciding on it.

What goes first is rarely the interesting work. It is frequently the grunt work and glue work that juniors historically handled as they learned their craft. The result is a leadership pipeline that looks full on paper but is structurally empty, and no pipeline means today's seniors become tomorrow's bottleneck.

Why nobody notices in time#

None of this shows up in a quarterly number. Junior employees still finish their work, and finish more of it than last year. What has gone is the learning that used to happen while they struggled with it, and nothing on any report measures that.

The bill arrives late and arrives all at once. Five years on there is a leadership bench that was never built, succession depends on hiring from outside, and the knowledge that lived in people who had done the work themselves has thinned without a single visible event.

Rebuilding the rungs#

  1. Redesign early-career roles around judgement, communication, and problem framing.
  2. Create structured rotations that expose people to decision contexts.
  3. Measure capability growth, not output.
  4. Protect human learning friction in the right places.
  5. Build new rungs intentionally.

The instinct to automate the junior work is understandable: it is the most visibly repetitive, the easiest to hand to a machine, the fastest efficiency win. But that work was never only production. It was the training ground where judgement was formed. Remove it without replacing it, and you save money this year at the cost of a capability you will need in five. The organisations that manage this well do not refuse to automate. They rebuild the ladder deliberately, so that the people who will lead them later are still learning to think now.

The same problem, described in 1983#

The strongest corroboration of this argument is forty-three years old and was written about process control rather than knowledge work. In Ironies of Automation, published in Automatica in 1983, Lisanne Bainbridge observed that automating a system removes the routine operation through which operators stayed practised, while leaving them responsible for the difficult cases. Her wording is precise: "the more advanced a control system is, so the more crucial may be the contribution of the human operator", and "by taking away the easy parts of his task, automation can make the difficult parts of the human operator's task more difficult."

That is the missing rungs, stated before the personal computer was widespread. It matters for two reasons. It shows the mechanism is a property of automation rather than a novelty of AI, which makes it considerably harder to dismiss as technophobia. And it means the aviation and process industries have four decades of hard-won practice in designing against it, which knowledge work has not yet bothered to read.

Independent field evidence#

Matt Beane reached the same problem from a different direction. In The Skill Code (Harper Business, 2024), built on his own ethnographic research including years observing robotic surgery, he documented surgical residents losing the hands-on time through which surgeons have always been made. The senior surgeon at the console works alone; the resident watches. The operation goes well. The training does not happen.

Beane calls what remains shadow learning: the informal, often rule-bending ways juniors scavenge the practice the official system has removed. It is a useful and uncomfortable finding, because it suggests the ladder does not simply vanish. It goes underground, becomes unevenly distributed, and rewards the confident and well-connected over the diligent.

And now, measured#

The 2026 Global AI Jobs Barometer from PwC put numbers on it. Analysing 2.4 million US entry-level jobs, it found that entry-level roles most exposed to AI are seven times more likely to require traditionally senior, human-intensive capabilities such as leadership, creativity and face-to-face interaction. Those roles grew 35 percent since 2019, while other entry-level roles shrank 10 percent.

Read that carefully, because it is not the story the headlines told. Entry-level work is being seniorised rather than simply disappearing. The junior tasks through which people used to climb are being automated away, and the expectation of senior judgement is arriving on the first day instead, before there has been any opportunity to build it. The ladder has not been shortened. Its bottom rungs have been replaced with a demand to already be at the top. That finding is US-only and drawn from job advertisements, which describe what employers ask for rather than what the work requires, but it is the closest thing to direct measurement this argument has.

The Big Four have started acting on this#

On 27 August 2026 the Financial Times reported that consulting firms are weighing requiring junior staff into the office more often, on the grounds that AI has made interpersonal skills more valuable. EY's UK head of consulting is quoted saying firms will "have to reduce flexibility, but in order to help the human skills", and that training in empathy, storytelling and leadership was dropped during the remote-working period while AI and technical skills were prioritised. KPMG is reinventing its in-person training. Deloitte and PwC began extra coaching for their youngest UK recruits in 2023 after finding weaker teamwork and communication than earlier cohorts.

This is the argument on this page arriving in the trade press with named executives attached. It is the strongest external corroboration this research has.

The diagnosis is right. The remedy does not follow from it. If juniors are weaker because the tasks that built judgement were absorbed, attendance does not restore them. A junior in an office while a model still writes the first draft has gained proximity, not repetitions. Presence and practice were bundled together for a century, so they are easy to confuse now that AI has separated them.

And the reporting contains no measurement. The 2023 cohort effects are attributed to pandemic lockdowns rather than to AI, EY as a firm restated its existing flexibility policy alongside its executive's comments, and everyone quoted has an interest in the answer. It evidences what large firms now believe and are doing. It does not evidence the mechanism.

What it looks like up close#

A year ago, in an Uber to the station in the pouring rain, my driver turned out to be a project manager in his fifties from a FTSE 100 company. He had waited six months for the official Copilot training and kept asking HR about the AI strategy, while younger colleagues used ChatGPT without asking. By the time the restructuring came, they had made themselves indispensable. He drives nights now. I wrote it up in The Missing Rungs: "AI isn't removing jobs. It's deleting the instructions for how to get one."

The rungs are missing at the bottom too. My eldest is at university. I have stopped giving her advice about industries and given her one question to ask at interview, from Talking Shop: "Who here is paid to let me be wrong?"

What a first job taught that a degree does not#

The rungs are not only technical. The first job most people have teaches four things at once and announces none of them, and the entry-level work now being automated is disproportionately the work that carried them. That argument, and what it means for anybody starting out, is set out at how do juniors become senior.

The progression this breaks#

Why removing early-stage work is not simply an acceleration is best explained by Hubert and Stuart Dreyfus in Mind Over Machine (1986). They describe expertise as a progression through stages, from rule-following novice to situational, intuitive expert, in which each stage depends on the accumulated experience of the one below. Intuition, in their account, is compressed experience rather than a shortcut around it.

If that is right, and forty years of naturalistic decision research broadly supports it, then removing the early stages does not speed up the climb. It removes it. You cannot arrive at stage five having skipped one through three, because stages one to three are what stage five is made of.

Measured in German manufacturing#

The most rigorous test of this argument comes from a different technology and a different continent. Dauth, Findeisen, Suedekum and Woessner, publishing in the Journal of the European Economic Association in 2021, used German administrative worker and plant records from 1994 to 2014 to trace what industrial robots actually did to careers.

Incumbent workers were largely fine. They kept their jobs and moved into new, higher-quality tasks inside their original plants. The cost fell somewhere else entirely: on young labour-market entrants, who shifted away from vocational manufacturing training towards university because the entry route had closed behind them.

That is twenty years of hard administrative data finding exactly this pattern. The damage from automation fell on skill formation, not skill possession, and it was invisible if you only looked at the people already doing the job. Generative AI is not an industrial robot and the analogy should not be pushed too far. But the mechanism has been observed at national scale before, which makes it considerably harder to dismiss as speculation about knowledge work.

September and October 2026: the consultancies count the rungs#

The autumn's consultancy reports put numbers on the pattern this page describes. Bain's survey of 155 software companies, published on 29 September 2026, found that junior engineers "used to make up one-third of teams on average, and now they're less than a fifth while senior ranks swell". Its verdict on the redesign so far: "Most companies have changed the easy things and left the hard ones untouched."

Deloitte's AI agents in the human workplace (10 September) built four scenarios rather than forecasts, and in one of them named the task this page argues for: "preserving the human judgment pipeline before it atrophies". Accenture's Future of Work report says one in five managers are already concerned that junior employees are missing hands-on experience, though the report does not give the source of that figure. BCG's study of deskilling found 53 per cent of senior executives citing the slower development of junior talent.

Meanwhile PwC's Hopes and Fears survey of 49,364 workers found access to learning and development falling from 59 to 51 per cent in a year, while AI use at work rose ten points to 64 per cent. The bottom rungs are thinning, and the formal training that might replace them is thinning too.

23 to 29 October 2026: nine years to grow one, three years to break the pipeline#

Two documents published a week apart put arithmetic on the shape this page describes. Ina Catrinescu's working paper of 22 October models a firm as two cohorts rather than one worker, and asks what automating codified work does to the pipeline that turns juniors into seniors. Because junior work is the most codified, demand for juniors falls first; the senior stock falls later and further, because nobody is replacing the experts who leave. In her baseline scenario junior demand drops about a fifth below its 2022 level before mostly recovering, the senior stock falls by about a quarter and later roughly halves, and the time to develop an expert inside the firm rises from about five years to about nine. A manager optimising over three years cuts the wage bill sharply while the firm's capacity on the work AI cannot do falls, and the cost falls on a different party from the one making the decision, which she calls an adaptability tax. She is explicit that two parameters are unmeasured and that the numbers are scenarios rather than forecasts. The part worth keeping is not the nine years but the three things she says a firm could count now: the unaided practice juniors get on work AI cannot do, the time seniors spend on such work, and the years it takes to grow an expert from within.

Niam Yaraghi reached the same place from research rather than from firms. Writing at Brookings on 29 October, after OpenAI released 722 machine-generated mathematical manuscripts of which three were withdrawn over a sign error, he argues that production has scaled and absorption has not: "Absorption does not scale." His account of why names the rung: "The people who can check frontier work are produced by doing frontier work." There is no consumption-only route to that competence, so paying a graduate student to do work an agent finishes in an afternoon looks indefensible to a department chair, and the capacity to verify disappears with the training that produced it. His two sentences are the sharpest statement of the asymmetry on this page: "The pipeline that produces those people takes a decade to refill. It takes about three years to break." He notes US National Science Foundation graduate fellowships falling from nearly 2,600 awards in April 2026 to an expected 1,500, against a budget request proposing a cut of just over 40 per cent.

The labour market supplied its usual evidence, and as usual it will not say why. Tom Haines at London Loves Business reported Adzuna's September figures on 27 October: advertised graduate vacancies at 7,826 in September, “down from 13,754 a year earlier”, a fall of 43 per cent, with 2.33 jobseekers for every advertised role and an average advertised graduate salary of £27,700. The article does not mention AI at all, and gives a different cause: “The contraction reflects weaker recruitment across professional services, technology and other sectors.” That is the caution this page has applied to every such figure, stated by the source itself. Around it, three professions described the mechanism in their own words. Lord Neuberger, a former president of the UK Supreme Court, told a conference in Kuala Lumpur that he expects “a real fall-off in recruitment of young lawyers” in the medium term, because AI-drafted pleadings, advice letters and research notes are often as good as a starting lawyer's and can be produced more quickly and cheaply, and he asked “how do we get the next generation of experienced senior lawyers”. The New Straits Times report could not be read at source and is relayed by Resultsense, which notes he did not predict it with certainty; the comparison with a starting lawyer's work is its paraphrase rather than his words. A LexisNexis survey written up by Karen Waldron for Artificial Lawyer on 29 October, a sponsored piece, found 75 per cent of legal professionals thinking juniors who use AI heavily may struggle to develop legal judgement, and recommended that supervisors assess how a junior evaluated the AI's answer rather than the document alone. And Dayforce's survey of 6,909 people on 26 October found 51 per cent saying AI is reducing early-career opportunities and 49 per cent that tasks which once built work experience are now handled by AI, which it names experience debt.

One rung was removed on the record, by a minister, as an improvement. Reporting HM Land Registry's processing times falling from 20 months in February 2023 to under 12 months by April 2026, Sam Trendall at PublicTechnology on 28 October quoted Baroness Taylor saying the registry "is also developing the use of artificial intelligence to undertake administrative tasks" so that skilled staff can be "focused on applications that are more complex or require legal judgement". As a service improvement it is one. Read as a pipeline, it removes the simple cases on which a new caseworker learns to recognise a hard one, and nothing in the announcement says what replaces them.

30 October to 5 November 2026: the rungs are rebuilt as checking, and a third party measures the thesis#

The week described the replacement rung in detail. It is a checking job. MaryLou Costa at BBC News on 4 November profiled graduates hired to manage AI agents: a 25-year-old "AI associate" at a PR agency whose team has built around 30 agents, and a 23-year-old consultant who said "progression is no longer based on your knowledge bank" and that he was "now taking on responsibilities which would usually take me four or five years to get to". PwC UK's Claire Reid told the BBC that "There will be no graduate job now that doesn't require a baseline of AI literacy", and then that "AI skills on their own are not going to help you in your career." PwC's 2026 AI Jobs Barometer, from 2.4 million American entry-level postings, found the entry-level roles most influenced by AI "seven times more likely to require traditionally senior-level skills"; those roles grew 35 per cent from 2019 and the rest shrank 10 per cent. Irina Anghel at Bloomberg, in a report carried by The Star on 5 November, gave the British figures: Adzuna says entry-level postings mentioning AI have almost tripled since the start of 2025 while junior vacancies are down a quarter over the same period; Indeed says one in six entry-level vacancies in exposed occupations now mentions AI, double a year ago, and that demand for AI skills "has only offset a third of the declines in other entry-level postings"; almost one million 16 to 24 year olds are neither working nor studying. Harvey Nash's Simon Crichton said firms are creating roles that hand juniors "the last 20% of the task" that AI leaves; Jo Drake, chief technology officer at THG Ingenuity, said of the review her team now does that "the man-marking of the agents, actually that's quite good for education."

The professions said what the rung used to teach. Daniel Hood, editor of Accounting Today, wrote on 5 November that "without that kind of training ground, the rising generation of accountants won't be trained", that staff are now expected to be ready for higher-level work "in, say, two years, as opposed to five or six or seven", and that "Deadline pressure and the hot breath of a client on your neck are not learning aids"; his remedy is "setting aside unbillable time for training, and creating real curricula". Grace Noto at CFO Dive on 4 November reported EY's Myles Corson asking how to develop people "when the traditional training ground and entry level rules where people earn their spurs are disappearing", and Sandra Oliver, who oversees a cohort of 130,000 assurance professionals, saying juniors will do "more of that judgment work" and professional scepticism earlier: "instead of teaching our staff like what to think, you're teaching them how to think." EY has turned an eight-week internship into a paid residency of eight to twelve months. Orianna Rosa Royle at Fortune on 30 October reported Cisco's president giving a 27-year-old former founder "unlimited tokens" to refactor 98 per cent of the company's legacy code within 24 months, and asking: "If you don't hire the early-in-career [workers], and you just let your workforce continue the way that they were, where do fresh ideas get injected?"

The cost of a missing rung was measured from the other end. Nick Lichtenberg at Fortune reported on 30 October that Goldman Sachs economist Pierfrancesco Mei had analysed more than 15,000 workers displaced between 2016 and 2025 and found that those displaced from AI-exposed jobs since 2025 took about a month longer to find work and were roughly 10 percentage points more likely to be unemployed a year later; workers of 35 and over took about five weeks longer than other displaced workers to return to full-time work, those under 35 about three. The effects are "comparable in magnitude" to earlier waves of technological change, whose damage Goldman calls "long-lasting"; Goldman adds that it is "still too early to assess the impact along all of these dimensions". In the Lords on 3 November, Lord Leong said for the government that "AI is automating repetitive tasks, and we take the risk to entry-level jobs seriously", restated the aim of 10 million people with practical AI skills by 2030 and a plan for 20,000 entry-level digital jobs a year, and added: "I fully accept that the labour market is failing too many graduates." A model from KIT and IBM Research, posted on 4 November, gives the rung its mechanism: "novices can become trapped at low skill levels, while experts may lose skill, even as AI-assisted outcome accuracy remains high", and only workers who value their future skills, or are paid for correct unaided work, climb until "independent work can become self-sustaining once workers acquire sufficient skill". It is a model, not a cohort.

And the phrase was used by someone else, in the same sense, with no reference to this page. InfoQ published on 30 October the recording and transcript of Alasdair Allan's QCon London 2026 talk, The Ladder is Missing Rungs: Engineering Progression When AI Ate the Middle. "As an industry, we are facing a fundamental structural problem. The ladder is missing rungs." Of payroll data showing employment of 22 to 25 year olds in exposed occupations falling while nobody was sacked, he said: "This is the missing rungs thesis measured in the labor market in real time." His close is this page's argument in an engineer's words: the ladder "is missing the mechanism that created the people who built the ladder". Allan arrived there independently and credits nobody, which is how an idea becomes vocabulary. What the week adds is that the replacement rung is review. Whether checking an agent's output builds the judgement that doing the work built is the question nobody has yet measured through a cohort: Drake says it is good for education, Hood says deadline pressure is not a learning aid, and both may be right about different firms. The decision an organisation controls is whether the last 20 per cent is designed as training, with a named person responsible for what the junior must become able to do unaided, or left as whatever the agents did not finish. The questions are at Rules Before Tools.

Further reading#

The full graded evidence is in the evidence base, and the wider map of the field, including Bainbridge, Beane and Dreyfus in full, is in the essential works.

The individual version of this is synthetic seniority; the practice version is the missed reps; the organisational accumulation is capability debt. See also will AI replace entry-level jobs and how humans learn with AI.

Development of the idea#

The term was introduced in the Box of Amazing essay The Missing Rungs: What Nobody Will Tell You About AI and Your Job on 21 September 2025. The argument was anticipated a year earlier in Critical thinking (10 November 2024) and developed alongside The Great Unbundling of Work (25 May 2025) and The Half-Life of Skills (8 June 2025). For the full dated record, see the timeline.

The older name for what is being lost is tacit knowledge: the knowledge that resists articulation, is acquired by doing, and passes through shared work rather than documentation. Its transmission route is the work being automated.

About this research#

Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company.

How this research works · Reviewed quarterly · Found an error? Tell me and it is corrected on the page.

Listen. Heard on Education Futures, SuperSkills: The 7 human skills AI can't replace (with Rahim Hirji). Heard on The Entropy Podcast, SuperSkills for the AI Age with Rahim Hirji.

Essay · SS-2025-001 · How I use AI here · 4 institutional surveys, 2 working papers, 2 compiled reviews and 5 of other kinds

Cite this page

Hirji, R. (2025). The Missing Rungs. The SuperSkills evidence base, SS-2025-001. https://thesuperskills.com/research/missing-rungs. Last reviewed 9 November 2026.

An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.

How citations and IDs work

In this hub

The named concepts

The vocabulary this research contributed, and what each term does.

Ask the evidence
What does the evidence actually show?What should our board be asking about this?Where does Rahim disagree with the consensus?
Bring this into your organisation

If this describes something happening in your teams, say so.

Keynotes, board sessions and advisory work, drawing on research across more than 200 organisations in 30 countries. Send me the date, who the audience is and the shift you need. I reply myself, within a working day.

Enquire

Topics and audiences  ·  All research

This is the part of the argument that hits differently with the people it is about. See AI keynote on early careers and graduate talent, and the full range of topics and audiences.

Reading this because it’s your problem?

Every week I write one essay on what AI is doing to judgement, with ten links worth your time. Box of Amazing has gone out since 2017, and 25,000 people read it. Free, and one click to stop.

You’ll get Box of Amazing weekly. One click in any issue stops it. How your email is used.

Working on this inside your organisation? Start with what you’re deciding

AI keynote speaker · London · Middle East · Advisory · Boards