"The Missing Rung in Practice" -- How the Big Four Are Cutting Their Own Ladder

Twin Ladder Casebook Series | Twin Ladder | February 2026

The Hook

She starts on a Monday in September. The graduate intake is smaller this year -- the firm does not say by how much in the orientation deck, but the cohort knows. There are fewer desks in the bullpen than the photos on the careers page suggested. The partner who delivers the welcome address mentions transformation four times in twelve minutes.

Her first assignment arrives by Thursday. In the years before, it would have been a bank reconciliation -- three days of cross-referencing ledger entries against statements, learning where the numbers hide, developing the instinct for where they do not add up. She would have made mistakes. A senior would have caught them, explained why, and sent her back to find the rest. By the end of the month, she would have understood the anatomy of a financial statement at the level that only comes from handling every bone.

Instead, she is given a laptop with an AI agent already running. The reconciliation is done. Her task is to review the output for anomalies. She reads through the results carefully, flags nothing, because she does not yet know what an anomaly looks like in this context. She has never built a reconciliation from the ground up. She is checking work she has never learned to do.

She does not know it yet, but she is standing on a ladder with the bottom rung removed.

The Story

The Billion-Dollar Bet

The Big Four accounting firms -- Deloitte, PricewaterhouseCoopers, Ernst & Young, and KPMG -- generated more than $212 billion in combined global revenue in 2024. They employ over a million professionals across every major economy. For decades, their business model has rested on a structural principle so fundamental that it is rarely examined: the pyramid.

The pyramid works because it is self-reproducing. Large cohorts of graduates enter at the base. They perform the high-volume, detail-intensive work -- the audits, the compliance checks, the data gathering -- that generates revenue at low marginal cost. In doing so, they learn. They absorb the tacit knowledge that turns competent analysts into trusted advisors. A fraction advance to the next tier, carrying that knowledge upward. The pyramid simultaneously generates profit and produces the next generation of partners. It has operated this way for a century.

In 2025 and 2026, all four firms are making a bet that they can replace the base of the pyramid with artificial intelligence and keep the structure standing.

EY: 150 Agents and Counting

Ernst & Young has deployed over 150 AI tax agents supporting 80,000 tax professionals across data collection, document analysis, and compliance -- a deployment designed to process more than three million tax deliverables and reshape thirty million tax processes. The firm invests more than $1 billion annually in AI platforms, with over 1,000 agents in development and more than 100 internal AI applications already operational. The investment is paying off in revenue terms: AI-related consulting brought in a 30 percent revenue increase in the 2025 financial year. EY has simultaneously reduced its graduate intake by 11 percent, from 1,800 to 1,600.

KPMG: The Workbench

KPMG launched Workbench in June 2025 -- a multi-agent AI platform built on Microsoft technology, described as the foundation for all client delivery. At launch, the platform contained 50 AI assistants with nearly a thousand in development. The investment is part of what the firm describes as a "multi-billion dollar" commitment to AI and agentic transformation. KPMG has implemented the steepest graduate cuts of any Big Four firm: a 29 percent reduction, from 1,399 hires to 942, with approximately 1,000 expected this year. The firm has also laid off around 4 percent of its U.S. audit workforce -- approximately 330 employees -- citing lower voluntary turnover and the need to align workforce size with market demands.

Deloitte: Zora

Deloitte unveiled Zora AI in 2025, an agentic platform built on NVIDIA AI that automates procurement, finance, human capital, and supply chain processes. The firm intends to use Zora internally to streamline its own expense management, targeting a 25 percent cost reduction and 40 percent productivity improvement. Hewlett Packard Enterprise, an early adopter, anticipates that Zora will reduce reporting production time by 50 percent. Deloitte has cut its graduate intake by 18 percent, from 1,700 to 1,400, and expects flat recruitment going forward.

PwC: The Quiet Contraction

PricewaterhouseCoopers has taken a different approach to communication but arrived at the same destination. Marco Amitrano, head of PwC's U.K. practice, publicly acknowledged in September 2025 that the firm is cutting entry-level roles, citing AI and struggling productivity as the primary drivers. In the United Kingdom, the firm hired 200 fewer entry-level staff, reducing intake to 1,300. But leaked internal documents revealed the larger picture: PwC plans to cut U.S. entry-level hiring by almost a third over the next three years. Tax and assurance associate hiring will fall by 32 percent between 2025 and 2028. In audit, the projected reduction is 39 percent. The firm spent nearly $1.5 billion on AI capabilities between July 2024 and June 2025 and has abandoned its previously stated goal of hiring 100,000 people by mid-2026.

The Warning From Inside

The implications have not gone unnoticed by those who built their careers within these firms. Alan Paton, formerly an equity partner in PwC's financial services division and now CEO of the Google Cloud consultancy Qodea, issued a direct warning: most structured, data-heavy tasks in audit, tax, and strategic advisory will be automated within three to five years, eliminating approximately 50 percent of roles. AI solutions capable of performing 90 percent of the audit process already exist, he stated. His concern extended beyond jobs to the business model itself: if clients can get answers instantaneously from a tool, they will increasingly question why they should pay consultants to deliver the same answers at premium rates.

Alongside the hiring reductions, a quieter shift is underway. Starting salaries across the top firms have been frozen, in some cases for three consecutive years. The Irish Times reported in late 2025 that top consultancies have frozen starting salaries as AI threatens the pyramid model. The economic logic is straightforward: if AI handles the work that graduates used to do, the market-clearing wage for new graduates falls. But the developmental logic runs in the opposite direction. If the firms are hiring fewer graduates into roles with less hands-on practice, the graduates who do enter face a longer, steeper climb to competence -- with less support, fewer mentors per cohort, and a narrower base of peers from whom to learn informally.

The pattern across all four firms is consistent. Invest billions in AI. Deploy hundreds of agents. Cut the graduate pipeline. Freeze starting salaries. The technology strategy is coherent. The workforce strategy is not.

Through the Twin Ladder Lens

The Level 0-to-1 Gap

Twin Ladder's Twin Ladder framework defines four progressive levels of AI competence. Level 0 is AI Literacy -- the baseline ability to critically evaluate AI-generated output. Level 1 is the Professional Twin -- mirroring individual roles with AI agents so that professionals compare, challenge, and learn from AI output while preserving their domain expertise. Level 2 is the Operational Twin. Level 3 is the Ecosystem Twin. The ladder is climbed, not skipped.

The Big Four are attempting to leap from no framework to Level 1 -- deploying Professional Twins at scale -- without ensuring that the professionals who must work alongside those twins possess the foundational competence that makes the relationship productive. They are deploying AI agents that perform reconciliations, draft compliance reports, analyze tax positions, and evaluate supplier contracts. They are then asking graduates, who have never performed these tasks manually, to supervise the output.

This is the structural flaw. The Missing Rung, as documented in Twin Ladder's white paper of the same name, is not merely a labour market problem. It is a competence architecture problem. You cannot critically evaluate work in a domain you have never practiced. You cannot identify the anomaly in a reconciliation if you have never built one. You cannot challenge an AI-generated tax position if you have never reasoned through the logic yourself. The supervision task presupposes the very expertise that the elimination of entry-level practice prevents from forming.

The graduate who reviews the AI agent's reconciliation output on her first Thursday is not operating at Level 1 of the Twin Ladder. She is operating at Level negative one -- exposed to AI output without the domain competence to assess it. She will approve work she does not understand. She will miss errors she has never been trained to recognize. And the longer this continues, the harder it becomes to close the gap, because uncritical reliance on AI output is itself a mechanism for competence erosion. The Competence Paradox, documented in Twin Ladder's companion paper Thirty Years Is Too Long, operates at the individual level: every month of passive AI reliance makes the professional less capable of the critical evaluation that responsible AI use demands.

The ladder is climbed, not skipped. The Big Four are asking people to stand on a rung that is not there.

The Pattern

The Protege Effect in Reverse

Twin Ladder's The Missing Rung white paper identifies a loss that extends in both directions when organizations eliminate entry-level roles. The visible loss is at the bottom: graduates lose the apprenticeship through which expertise forms. The hidden loss is at the top.

The Protege Effect, documented across multiple studies (Nestojko, Bui, Kornell, and Bjork, 2014; Fiorella and Mayer, 2013; Chase, Chin, Oppezzo, and Schwartz, 2009), demonstrates that people who teach material develop deeper, more durable understanding than people who merely study or review it. In controlled experiments, participants who prepared to teach outperformed those who prepared to be tested -- even when no teaching actually occurred. When actual teaching took place, the effect was stronger still. Teaching is not a distribution mechanism for existing knowledge. It is a generation mechanism for deeper knowledge.

An audit partner who explains to a graduate why a particular reconciliation approach is flawed is not merely training the graduate. She is re-examining her own assumptions, testing her reasoning against fresh questions, and updating her mental model of the domain. When that graduate is replaced by an AI agent, the partner still reviews the output. But reviewing is not teaching. Evaluation is not generation. The cognitive exercise that the research shows produces deeper understanding -- the act of organizing knowledge into a coherent explanation for another person -- disappears.

The pyramid was not merely a business model. It was a learning architecture. The teaching relationship between seniors and juniors served a function that neither party fully recognized: it kept the seniors sharp. Eliminate the juniors, and the seniors lose the feedback loop that maintained their judgment. The pyramid reproduced itself. The obelisk that replaces it -- tall, narrow, with a diminished base and an increasingly isolated senior tier -- does not.

The insurance industry provides a parallel that should concern the Big Four directly. Underwriting is being transformed by AI at a pace that has prompted 73 percent of underwriters and 82 percent of actuaries to report that they lack the skills they will need for the future. The training pipeline is narrowing precisely as the demands on senior judgment are increasing. The industry is automating the tasks through which the next generation of underwriters was supposed to learn, while simultaneously requiring that generation to supervise AI output in a domain they have not mastered. The pattern is identical.

The Lesson

IBM's Counter-Thesis

In February 2026, IBM announced that it would triple entry-level hiring in the United States -- a decision that runs directly counter to the Big Four strategy. Nickle LaMoreaux, IBM's chief human resources officer, articulated the reasoning with unusual clarity: cutting early-career recruitment creates a longer-term scarcity of mid-level managers and experienced workers. The companies that will be most successful in three to five years, she stated, are those that doubled down on entry-level hiring in this environment.

IBM did not ignore AI. It rewrote its entry-level roles to account for AI fluency. Software engineers spend less time on routine coding and more time on customer interaction. Human resources staff work on intervening where chatbots fail rather than answering every routine question. The entry-level position has been redesigned, not eliminated. The rung has been rebuilt, not removed.

This is the Twin Ladder approach in practice, whether IBM uses the language or not. Juniors learn by evaluating AI output, not by being replaced by it. They build domain competence through structured comparison -- seeing what the AI produces, understanding where it falls short, and developing the judgment to know the difference. They become the next generation of supervisors, managers, and partners with the expertise required to direct AI rather than defer to it. And in doing so, they create the teaching relationships through which their seniors remain sharp.

The Big Four are cutting costs today by narrowing the base of the pyramid. IBM is investing in capability tomorrow by widening it. One of these strategies produces a short-term margin improvement. The other produces the organizational competence to sustain margin over the next decade.

The difference is not ideological. It is structural. IBM recognized that entry-level roles are not simply a cost line. They are the mechanism through which an organization reproduces its capability. Cut the entry level, and you save money in year one. By year five, you face a mid-level talent gap that no amount of lateral hiring can close -- because the domain knowledge that makes mid-level professionals valuable was built during the entry-level years you eliminated.

The Big Four face this timeline on a larger scale. If audit associate hiring falls by 39 percent over the next three years, as PwC's internal documents project, who will be the audit managers of 2032? If KPMG's graduate intake has dropped by 29 percent, where will the partners of 2040 come from? The pyramid does not merely employ graduates. It produces the seniors who employ them in turn. Break the cycle at the bottom, and the effects propagate upward with a delay that makes them invisible until they are irreversible.

The question is not whether AI will transform professional services. It will. The question is whether the firms that built their reputations on human judgment will retain the capacity to exercise it.

Monday Morning Question: If your firm is deploying AI agents to perform the tasks that used to train your junior professionals, what is your plan for producing the senior professionals of 2035?

Sources

"Big Four giant EY is all in on AI -- and it's paying off." Yahoo Finance, 2025. https://finance.yahoo.com/news/big-four-giant-ey-ai-193040548.html
"KPMG launches KPMG Workbench: a multi-agent AI platform." KPMG, June 2025. https://kpmg.com/us/en/media/news/kpmg-launches-kpmg-workbench-a-multi-agent-ai-platform.html
"Deloitte Unveils Zora AI, Agentic AI for Tomorrow's Workforce." Deloitte, 2025. https://www.deloitte.com/us/en/about/press-room/deloitte-unveils-zora-ai-agentic-ai-for-tomorrows-workforce.html
"The Big Four are shrinking their grad schemes, with one company cutting roles by 30 per cent." The Tab, July 2025. https://thetab.com/2025/07/10/the-big-four-are-shrinking-their-grad-schemes-with-one-company-cutting-roles-by-30-per-cent
"PwC's U.K. chief admits he's cutting back entry-level jobs." Fortune, September 2025. https://fortune.com/2025/09/08/pwc-uk-chief-cutting-entry-level-junior-gen-z-jobs-ai-economic-headwinds-like-amazon-salesforce/
"A Big 4 Firm Is Cutting Back on Entry-Level Hiring, According to a Leaked Slideshow." Entrepreneur, 2025. https://www.entrepreneur.com/business-news/pwc-reducing-entry-level-hiring-changing-processes/496198
"Former PwC Partner Issues AI Warning to Big Four." Unity Connect, 2025. https://unity-connect.com/our-resources/news/ai-targets-big-four-warns-former-pwc-partner/
"IBM is tripling the number of Gen Z entry-level jobs after finding the limits of AI adoption." Fortune, February 2026. https://fortune.com/2026/02/13/tech-giant-ibm-tripling-gen-z-entry-level-hiring-according-to-chro-rewriting-jobs-ai-era/
"IBM To Triple Entry-Level Hiring, Warns AI-Driven Hiring Cuts Could Hollow Out Future Leadership." AllWork, February 2026. https://allwork.space/2026/02/ibm-to-triple-entry-level-hiring-warns-ai-driven-hiring-cuts-could-hollow-out-future-leadership/
"Big Four slash graduate jobs as AI takes over entry-level tasks." Scottish Financial News, 2025. https://www.scottishfinancialnews.com/articles/big-four-slash-graduate-jobs-as-ai-takes-over-entry-level-tasks
Twin Ladder. The Missing Rung: How AI Is Dismantling the Career Ladder -- And Why Your Best People Are Paying the Price. February 2026.
Twin Ladder. Thirty Years Is Too Long: The Competence Paradox. 2026.