The ladder was not a metaphor. It was a mechanism.
The career ladder is a phrase people use loosely. It is worth being precise about what it actually described, because you cannot think clearly about what is replacing it until you understand what it was doing in the first place.
In 1992, economists Robert Topel and Michael Ward documented something specific about how wages grow in the early stages of a career. The average American worker, they found, has seven employers in the first ten years of working life.[1] That churn was not dysfunction. It was the mechanism. About 60% of the wage growth in the first decade of a career comes not from raises within a firm but from moving between firms toward better-paying employers. You take a job, you learn, you get a better offer, you move. Repeat. That is how it was supposed to work.
The ladder, in other words, was not primarily about loyalty and promotion within a single company. It was about a functioning market for labor that allowed young workers to trade up over time. The first job did not need to pay well. It needed to teach something that made you worth more to the next employer. You did the rote work, you absorbed how the industry functioned, you developed judgment, and eventually you were doing work that required judgment rather than rote execution. The economics of that system depended on employers being willing to pay for the rote work phase, essentially subsidizing the development of talent they might not keep.
Brookings Institution and Opportunity@Work formalized this into a three-tier framework that is worth understanding precisely.[2] They call the tiers Origin, Gateway, and Destination.
Over the past decade, more than 23 million US workers without four-year degrees climbed this ladder from Origin through Gateway into Destination roles.[2] The ladder was not just a metaphor for people with degrees and corporate ambitions. It was the primary mechanism of economic mobility for tens of millions of workers who built their skills through work rather than formal education.
The bottom rungs are disappearing. The mechanism is breaking.
The Topel and Ward mechanism depends on employers being willing to hire workers into the rote-work phase. The economic rationale for doing that is collapsing. When an AI can generate an SQL query, summarize a legal brief, debug a code block, draft a first-cut memo, or process a customer service inquiry at near-zero marginal cost, the business case for paying a junior worker $60,000 a year to do the same work mostly disappears. Not because the junior worker does it poorly. Because the junior worker doing it well is no longer the cheapest way to get it done.
This phenomenon has a name in the research literature. It is being called the automation of "drunt work" — the digital grunt work that formerly comprised the bottom rung of the knowledge work ladder.[3] Summarizing meetings. Cleaning data. Drafting memos. Processing information. Preparing first-cut analyses. These were not glamorous tasks. But they were the tasks that gave people their first grip on the ladder. When they disappear, the climb does not just get harder. For many people, the starting point disappears entirely.
"The ladder is not broken — it is just being replaced with something that looks a lot flatter."
Heather Doshay, partner at SignalFire[4]She is right that it is flatter. She may be too optimistic about how smoothly the replacement happens. A flatter structure means fewer rungs, which means fewer people can climb it, which means the gains from reaching the top are concentrated in fewer hands. The flatness is not neutral. It is a distributional outcome.
The gateway problem: why the middle matters most
The part of this story that gets the least attention is not what happens at the bottom rung. It is what happens to the Gateway tier — the middle roles that connected entry-level work to professional careers.
Brookings identifies Gateway occupations as the most AI-exposed tier in the entire ladder.[2] Administrative assistants, customer service representatives, bookkeeping clerks, junior coordinators — these are the roles that did two things simultaneously. They paid better than Origin roles, so they attracted workers who were ready to move up. And they required enough skill accumulation that workers who held them for a few years came out the other side genuinely more capable, ready for Destination roles in accounting, management, paralegal work, and financial services.
When you automate the Gateway tier, you do not just eliminate a layer of jobs. You cut the bridge between where most people start and where they might hope to end up. The Destination roles do not disappear. The route to them does.
Of the roughly 70 million US workers without four-year degrees, 15.6 million work in jobs in the top quartile of AI exposure — 43% of all US workers most exposed to AI disruption. The Gateway roles they occupy are the most concentrated: administrative assistants, customer service representatives, and bookkeeping clerks account for the largest share.
Brookings identifies the Northeast and Sun Belt as most geographically exposed. Florida metros are particularly concentrated: Palm Bay at 35.5% of non-degree workers in highly AI-exposed Gateway roles, Orlando and Tampa at 32.2%. These are not abstractions. They are specific communities whose economic mobility infrastructure is being removed.[2]
The deeper problem is that Gateway jobs did not just provide income. They provided the specific skill development — exposure to organizational systems, client relationships, financial processes, institutional knowledge — that made workers worth more to the next employer. Brookings is clear on this: "Workers are more likely to move into higher-paying occupations that share underlying skill similarities with their current job."[2] Remove the Gateway, and you do not just eliminate a job category. You break the skill accumulation pathway that made upward mobility possible.
The data nobody is comfortable with
There are two competing narratives about AI and the career ladder right now. One says this is primarily a cyclical hiring freeze caused by interest rate hikes and post-pandemic correction. The other says AI is fundamentally restructuring which labor markets will hire entry-level workers at all. Both narratives have data behind them. The honest reading is that both are true simultaneously, and the interaction between them is worse than either alone.
The finance and information services number deserves particular attention. These two sectors were, for decades, the canonical on-ramps for college graduates into professional careers. Investment banks, consulting firms, media companies, tech companies: they absorbed enormous numbers of entry-level workers and ran them through the rote-work phase before promoting the best ones. Finance and information services were adding 44,000 jobs per month before the pandemic. They are now shedding 9,000 jobs per month.[5] That is not a cyclical blip. That is a structural reversal of one of the primary talent pipelines in the knowledge economy.
The underemployment number is equally striking. 42.5% of recent college graduates are working jobs that never required a degree.[5] Almost half the people who just spent four years and six figures on a degree are either barista-ing, retail-ing, or otherwise working outside their field of training — not because there is something wrong with them, but because the entry points to their intended careers have frozen or disappeared. The degree bought them into a game where the first move is unavailable.
And the AI attribution number from Challenger is worth holding carefully. In March 2026, 25% of all announced US layoffs cited AI as a primary reason — up from 10% in February and just 5% for all of 2025.[7] The trend line is hard to dismiss even accounting for AI-washing, the practice of attributing financially motivated cuts to AI because it sounds more inevitable and less negotiable than "we want to reduce headcount."
Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen's 2025 working paper found a 16% relative decline in employment for early-career workers aged 22 to 25 in the most AI-exposed occupations since ChatGPT's public release in November 2022.[8] Brynjolfsson reiterated the finding at a Stanford panel in March 2026, with more granular estimates: approximately 20% decline for entry-level software developers, approximately 15% for call center workers. Mid-career workers in the same occupations are doing fine. Senior workers are doing well. It is specifically the entry point that is contracting.
The paper's title is deliberate. Canaries in the coal mine die first. Their deaths are early warnings, not the primary event. The question the paper raises but does not answer is whether the workforce entering now is a canary or a casualty.
Why this is happening now and not five years ago
The technology that can automate drunt work has improved dramatically since 2022. But capability is not the only explanation for the timing. Three things converged.
The capability threshold crossed. Before large language models reached their current level, automating knowledge work tasks required custom software built for specific workflows. That was expensive enough that the economic calculus often favored keeping the junior worker. Current AI tools can handle a wide variety of unstructured text tasks — summarizing, drafting, researching, classifying, extracting — with minimal custom setup. The per-task cost dropped by orders of magnitude.
The labor market tightened at exactly the wrong time. The "low-hire, low-fire" environment that has characterized 2025 and 2026 means employers are not expanding. When hiring is flat, organizations facing a choice between replacing a departing junior worker and absorbing their workload with AI tools are making that calculation for the first time. The answer is increasingly: do not replace them. This is happening at the hiring decision level, not the layoff level, which is why the aggregate employment data does not fully capture it yet. The ladder is losing rungs through attrition and hiring freezes, not primarily through mass layoffs.
The post-pandemic over-hire correction accelerated the transition. Tech companies that over-hired in 2020 and 2021 cut aggressively in 2022 and 2023. Those cuts disproportionately hit junior roles. Many of those roles have not been refilled, not because the companies shrank permanently, but because the work was redistributed to AI tools and to senior employees with higher output per hour. The correction provided cover for a structural change that might otherwise have been more visible and more contested.
What replaces the ladder
The honest answer is that nobody has fully figured this out yet. "We still haven't figured that out" is the most credible thing Thomas Davenport at Babson said about it, and he has been studying this longer than most.[9] But the outlines of several replacement structures are visible, and they are worth thinking through honestly rather than optimistically.
The most thoughtful framing of this I have seen comes from Brookings: the question is not which jobs disappear, but how AI reshapes the pathways that connect entry-level work to higher-wage opportunities.[2] Pathways are not jobs. They are sequences of roles through which skills accumulate and economic mobility happens. A pathway can degrade even when individual job categories survive. The Gateway roles may still exist in smaller numbers. But if there are not enough of them to serve as a functional bridge for the workers who need to cross it, the system is broken even if the official job classification still exists.
"The central labor market challenge of AI may not be as simple as which workers or jobs are most disrupted, but rather how AI reshapes the career pathways that connect entry-level and low-wage workers to higher-wage opportunities."
Brookings Institution, April 2026[2]What this means for you specifically
The implications depend heavily on where you are in your career, not just what industry you are in.
If you are early-career or entering the workforce now
The honest advice is to compress the drunt-work phase as aggressively as possible using AI tools, rather than expecting to be paid to do it the slow way. If the economic rationale for hiring someone to do rote knowledge work is collapsing, the response is not to do the rote work compellingly. It is to demonstrate from the start that you can operate at the level above it. Build a public portfolio of outputs rather than waiting to be given chances. The Topel and Ward wage-growth mechanism — move between firms toward better offers — still works if you have demonstrated outputs that justify better offers. It is much harder to access if your first three years consist of jobs that do not produce anything visible.
If you are mid-career in an exposed field
The demand compression risk is real even if your specific role is not directly automatable. The organization around you is shrinking its junior base, which means the talent pipeline feeding your level is thinning, which means promotional paths are narrowing and firms are hiring experience rather than developing it. The practical response is to make your judgment, relationships, and context-dependent knowledge as visible as possible — both internally and externally. The workers who will fare best in a thinner market are the ones who are known beyond their current employer, not just valued within it.
If you are an employer or manager
Davenport's question deserves a direct answer from you: if you stop hiring entry-level workers now, where are your senior workers coming from in eight years? The organizations making this trade are capturing short-term efficiency gains at the cost of long-term talent pipeline health. That is a reasonable trade to make knowingly. It is a dangerous trade to make by default, because the efficiency gains are immediate and visible, and the pipeline damage is slow and invisible until suddenly it is not.
If you are a non-degree worker in a Gateway role
The Brookings data on this is specific and uncomfortable. Your role, not the top of the ladder, is the most AI-exposed tier. The Gateway occupations that served as the mobility mechanism for tens of millions of workers are the ones being automated most aggressively. The physical-world pivot, while culturally undervalued, may represent genuine stability that Gateway roles no longer do. That is not a comfortable conclusion. It is an honest one.
Sources and references
- Topel, R. & Ward, M. (1992) — "Job Mobility and the Careers of Young Men." Quarterly Journal of Economics. The foundational research documenting that the average American worker has seven employers in the first decade, and that 60% of early-career wage growth comes from moving between firms rather than raises within them.
- Brookings Institution / Opportunity@Work — Heck, J. & Muro, M. "How AI May Reshape Career Pathways to Better Jobs." April 2026. Introduces the Origin / Gateway / Destination framework. Documents that 23 million non-college workers climbed this ladder over the past decade and that Gateway jobs are the most AI-exposed tier. brookings.edu →
- Rezi / labor market research (2026) — "The Crisis of Entry-Level Labor in the Age of AI (2024–2026)." Introduces the "drunt work" framing for the digital grunt work that formerly comprised the bottom rung of the knowledge work ladder. Also documents the 46% decline in UK tech graduate roles in 2024 and the 1.6% projected increase in Class of 2026 hiring. rezi.ai →
- CNBC / SignalFire (Heather Doshay) — "AI is not just ending entry-level jobs. It's the end of the career ladder as we know it." September 2025. Includes the Revelio Labs data showing 35% decline in US entry-level postings since January 2023 and the Doshay "flatter" quote. cnbc.com →
- governance.fyi / BLS / New York Fed — "Marc Andreessen Is Right That AI Isn't Killing Entry-Level Jobs. Interest Rate Hikes Are. And That's Not Even the Worst Part." April 2026. Synthesizes BLS data (37-year high in new entrant unemployment, finance/info services shedding 9K jobs/month vs pre-pandemic +44K), NY Fed underemployment data (42.5% for recent graduates Q4 2025), and the Topel-Ward wage mechanism. governance.fyi →
- Anthropic Economic Index 2026 — Published March 2026. Documents 3–16% employment decline for workers aged 22–25 in most AI-exposed roles, attributing the decrease primarily to a slowdown in hiring rather than an increase in layoffs. lacepartners.com (summary) →
- Challenger, Gray & Christmas — March 2026 Job Cut Report. AI cited as primary reason for 25% of all US layoffs in March 2026, up from 10% in February and 5% for full-year 2025. 4cornerresources.com (summary) →
- Brynjolfsson, E., Chandar, B. & Chen, R. (2025) — "Canaries in the Coal Mine." Working paper. Finds 16% relative employment decline for early-career workers (aged 22–25) in most AI-exposed occupations since November 2022. Reiterated by Brynjolfsson at a Stanford panel, March 2026, with estimates of ~20% for junior software developers and ~15% for call center workers.
- Babson College / Thomas Davenport — "AI, Jobs, and Uncertainty: A Leading Expert Weighs In." March 2026. Source of the "if companies don't hire entry-level workers today, how do you get experienced workers tomorrow?" quote. entrepreneurship.babson.edu →
- Goldman Sachs Research — "How Will AI Affect the Global Workforce?" August 2025. Documents that unemployment among 20–30 year olds in tech-exposed occupations rose nearly 3 percentage points since early 2025, notably higher than same-aged peers in other fields. goldmansachs.com →
- LACE Partners (2026) — "The End of the Graduate Career Ladder? How AI Is Reshaping Early-Career Talent." Synthesizes Anthropic Economic Index with broader graduate hiring trends and talent pipeline implications for HR leaders. lacepartners.com →
Frequently asked questions
Is the career ladder really ending, or is this just a temporary hiring freeze?
Both are happening simultaneously, which makes it harder to read. There is a genuine cyclical component: the post-2022 interest rate environment tightened hiring, the post-pandemic over-hire correction hit junior roles hardest, and the "low-hire, low-fire" market is devastating for anyone trying to enter. Those cyclical forces will ease. But underneath them, there is a structural change: the economic rationale for hiring workers into the rote-work phase of their development is weakening as AI tools handle that work at lower cost. The cyclical recovery will not fully restore the structural loss. Entry-level volumes may recover somewhat when the macro environment improves. They will not return to pre-2022 levels in the most AI-exposed fields.
Who is most at risk?
The Brookings data is specific on this.[2] The most at-risk group is non-degree workers in Gateway roles — administrative assistants, customer service representatives, bookkeeping clerks — who relied on those roles as the bridge to higher-wage work. 15.6 million US workers without degrees are in highly AI-exposed jobs. Among recent graduates, those entering finance, information services, and tech face the most compressed entry-level market. Geographically, Florida metros, the Northeast, and Sun Belt cities with high concentrations of administrative and clerical Gateway jobs are most exposed.
What about the argument that AI will create new jobs to replace the ones it destroys?
The historical argument that technology creates as many jobs as it destroys is true in the aggregate over long time horizons. It is insufficient as a response to what is happening now for three reasons. First, the new jobs tend to require different skills, in different locations, after a transitional gap that can last years. Second, the new jobs are often not at the entry level — they tend to require either technical expertise or significant experience. Third, the specific mechanism being disrupted, the paid training ground where workers develop judgment by doing rote work, is not automatically recreated by new job categories at the top. New jobs at the destination tier do not solve the problem of how workers get there.
What should someone entering the workforce right now actually do?
Three things that are probably worth doing regardless of how this plays out. First, compress the drunt-work phase: use AI tools to produce at a level above what your experience technically justifies, rather than waiting to be given the chance to demonstrate it the slow way. Second, build a visible portfolio of outputs rather than relying on job titles and tenure as your primary credential — the Topel and Ward wage-growth mechanism still works if you have demonstrated something worth paying for. Third, take the physical-world options more seriously than your educational background probably primed you to. Healthcare support, skilled trades, and infrastructure work are genuinely less AI-exposed and are seeing real supply compression as demand holds steady.
What should employers do?
The ones worth working for are asking Davenport's question seriously: if we stop developing junior talent now, where are our senior people in eight years?[9] The organizations building durable talent pipelines in 2026 are treating early-career development as a deliberate investment, designing AI apprenticeship structures where juniors direct and audit AI outputs rather than just execute rote tasks, and accepting that the paid-learning-curve model has changed but not disappeared. The efficiency gains from not hiring juniors are real and immediate. The pipeline damage is slow and invisible until it is not.
More honest thinking about AI and work
This piece is part of an ongoing series applying long-arc, practitioner-grounded analysis to questions the mainstream discourse is either avoiding or answering badly.
Read more essays →