If AI Eliminates the Bottom of Your Org Chart, Who Do You Promote to Manager in 5 Years?

Professional services firms, law firms, consultancies, agencies, built their business model on a simple geometry: a wide base of junior staff supporting a narrow group of partners and senior advisors. That base always did two jobs at once. It supplied the billable hours that made the leverage model profitable, and it quietly supplied the raw material, in the form of mistakes made and corrected, from which senior judgment eventually got built. AI is now automating exactly the tasks that base performed, and it is doing so faster than most firms have updated talent systems that decide who gets promoted. 

So, the question is not really whether AI will change the shape of these organizations. That is already underway, and it is increasingly well documented. The harder question is what happens to the pipeline that used to turn junior staff into senior judgment once the junior tier that trained them has been thinned out. 

The Shape of the Organization Is Already Changing 

EY’s 2026 research put a name to what is changing. The pyramid is shifting to a diamond. AI handles the transactional, rules-based tasks at the base, humans move into judgment intensive roles in the middle and top tiers, and mid-level talent becomes the focus of both quality control and productivity (Memani et al., 2025). This is already visible in hiring data. 

Economists at Stanford’s Digital Economy Lab used payroll records covering millions of workers to measure it directly. Employment for workers between the ages of 22 and 25 in the most AI-exposed occupations now sits roughly 19% below where it would be had those workers kept pace with similarly aged peers in less exposed fields, and the gap is being driven mainly by reduced hiring, not layoffs (Brynjolfsson et al., 2026). The young people affected are not losing jobs. They are just not getting them in the first place. 

What the Base Was Actually Doing 

It is tempting to treat this as a simple efficiency story: routine work gets automated, cost goes down, and the people who used to do that work move on to something else. That framing misses what the routine work was actually producing. In auditing, engineering, healthcare, and financial services, junior professionals have historically built expertise through apprenticeship, performing routine tasks, making mistakes, receiving feedback, and gradually developing the pattern recognition required for complex judgment. The junior auditor working through a high volume of financial records is not merely completing a routine task: she is learning from mistakes and gradually developing the case library and pattern recognition that later enables anomaly detection. 

Some audit firms have already automated up to 40% of junior level work, including testing controls, transaction sampling, and anomaly investigation (PCAOB, 2023). The short-term efficiency gains are real, but the longer-term effect on capability is a lot harder to see, and it does not show up in standard return on investment calculations. Ericsson, Krampe, and Tesch-Römer’s (1993) research on deliberate practice helps explain why: expertise requires structured engagement with a domain’s full complexity, including feedback that lets a practitioner calibrate judgment against real outcomes, and routine procedural work has historically supplied exactly that foundation. 

Stanford’s payroll data adds an empirical layer to the argument. The employment decline is concentrated in roles that depend on codified knowledge, the kind AI can reproduce, while experienced workers in those same occupations, whose value rests on tacit judgment built over years, show no comparable decline (Brynjolfsson et al., 2026). Firms are not losing their senior people yet. They may be quietly losing the conditions that used to produce the next generation of them. 

Apprenticeship Was Never Just Training 

AI automation by itself affects judgment formation. Combined with remote work and skills-based staffing, it reshapes the entire apprenticeship ladder. Apprenticeship is the practice of early career staff working at the feet of supervisors. This served as both quality control and talent development. Managers and supervisors explain misconceptions, diagnose flawed reasoning, share mental models, scaffold progressively complex work over time, and provide career advice that helps propel novices up the organization chart. 

Replicating what apprenticeship accomplishes remotely is non-trivial and takes deliberate design. Proximity alone will not do it. Losing a career mentor is a reliable predictor of attrition (Payne & Huffman, 2005). Skills based staffing can compound the problem by converting long-term working relationships into short term project assignments, which may push staff toward lateral movement rather than vertical advancement, trading what used to be a career ladder for something closer to a ‘career lattice’. 

What happens in half a decade or so?1 

The deeper problem is not that fewer people are getting promoted. It is that the systems firms use to decide who is ready for promotion were built on an assumption AI has quietly broken. Those systems assumed early career years would be developmentally dense, meaning the years themselves would reliably produce judgment as a byproduct of volume (Durham, Wallace, & Kupetis, 2026). Break that assumption without changing the promotion timeline, and firms end up promoting people into roles that need a level of judgment their compressed formation never had the chance to build. 

That gap is hard to spot, because it does not create outliers. It moves the whole cohort at once, so a practitioner whose AI compressed early career has produced less developed judgment than the role requires still looks normal relative to peers who went through the same compression (Durham et al., 2026). One industry commentator recently put the professional services version of the question plainly: how do future partners develop the depth the job requires if machines are doing much of the foundational work that used to build it (O’Dowd, 2026)? 

Not Everyone Agrees Yet 

The evidence is not uniform, and it is worth saying so. McKinsey, BCG, and Bain froze US undergraduate starting pay for a third straight year heading into 2026, a signal consistent with a shrinking base (Financial Times, 2025). McKinsey’s own North American chair pushed back publicly around the same time, announcing plans to hire 12% more graduates in 2026 and arguing the firm still needs the same level of intellect and the same pace it always has (as cited in Financial Times, 2025). 

Big Law tells a similar, mixed story. Associate starting salaries kept climbing through 2026 even as firms openly debated what AI means for their operating model, largely because none of them have yet been willing to abandon their leveraged pyramid structure (Bloomberg Law, 2026). The pyramid is not collapsing everywhere at once. It is being tested unevenly, and the firms still hiring the way they always have simply have not had to answer the question, at least not yet. 

Building What Used to Happen by Accident 

If experience does not accumulate automatically through repetition anymore, it has to be designed on purpose. Bain’s recent analysis of operating models for the AI era puts it directly: when execution gets automated, onboarding, training, and career paths need to evolve, because experience now has to be designed rather than assumed (van Dijk et al., 2026). That is a different job than most firms have ever had to do on purpose, because for most of the history of professional services, the work itself did the designing for them. In short, in an AI world, expert judgment and critical thinking become THE power skills – the problem is most firms only teach these skills by accident, and AI-automation promises to take much of that away. 

McKinsey’s own research on early career talent offers a workable starting point, built around three shifts. Firms need to change whom they hire, screening for judgment under ambiguity rather than speed of task execution. They need to change what new hires actually do, building deliberate rotations that pair junior and senior staff on real problems instead of routing junior staff toward whatever AI has not absorbed yet. And they need to change how mentorship works, treating it as a structural requirement rather than a nice to have that only happens if a manager finds the time (Durth et al., 2026). 

Who Fills the Middle 

Which brings us back to where we started. If the base thins, where do the people who populate the thick middle come from, if the pipeline from below is constricted? And just as important, how do the fewer people who get promoted develop the expertise they need to review AI output for accuracy, when the work that used to build that expertise got automated out of their early careers? 

For professional services firms built on billable hours, this is not an abstract talent question. It is a direct threat to the leveraged billing model that funds the whole structure. McKinsey’s own researchers put the stakes plainly: organizations that recognize this problem early will shape the next generation of leaders, and those that do not may find themselves without one (Durth et al., 2026). The firms that figure out how to build judgment on purpose, instead of hoping it keeps forming by accident, are the ones that will still have someone to promote in five years. 

Part of the answer requires a reinvestment in some of the operational efficiencies gained by automation, towards a learning transformation that recalibrates formal education systems towards deliberate practice and judgment formation. The more challenging part, the ultimate sustainability of diamond shaped org charts, may require a rethinking of recruiting practices, job architectures, staffing models, career paths, compensation structures – and in the case of some firms, revenue models. 

References 

Bloomberg Law. (2026, June 11). Law firm salaries climb despite AI promise of lower costs. https://news.bgov.com/antitrust/law-firm-salaries-climb-despite-ai-promise-of-lower-costs  

Brynjolfsson, E., Chandar, B., & Chen, R. (2026). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence (Rev. ed.). Stanford Digital Economy Lab. https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/  

Durham, L., Wallace, G., & Kupetis, T. (2026). The Judgment Void: How AI is dismantling the one human capability it cannot replace. St. Charles Consulting Group. The Judgment Void – St. Charles Consulting Group 

Durth, S., Maor, D., & Underwood, S. (2026, May 18). Rethinking early-career talent in the agentic organization. McKinsey & Company. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-organization-blog/rethinking-early-career-talent-in-the-agentic-organization  

Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100(3), 363–406. https://doi.org/10.1037/0033-295X.100.3.363  

Financial Times. (2025, December 1). Top consultancies freeze starting salaries as AI threatens ‘pyramid’ model. Reprinted in The Irish Times. https://www.irishtimes.com/business/2025/12/01/top-consultancies-freeze-starting-salaries-as-ai-threatens-pyramid-model/  

Memani, R., Makhija, M., Balaji, H., Pandharkar, R., Gupta, R., Sharma, G., Patni, J., & Dandavate, Y. (2025). Is India ready for agentic AI? The AIdea of India: Outlook 2026. EY. https://www.ey.com/content/dam/ey-unified-site/ey-com/en-in/insights/ai/documents/is-india-ready-for-agentic-ai-the-aidea-of-india-outlook-2026.pdf  

O’Dowd, J. (2026, January 7). AI quietly rewrites professional services firms’ talent model. Bloomberg Law. https://news.bloomberglaw.com/tech-and-telecom-law/ai-quietly-rewrites-professional-services-firms-talent-model  

Payne, S. C., & Huffman, A. H. (2005). A longitudinal examination of the influence of mentoring on organizational commitment and turnover. Academy of Management Journal, 48(1), 158–168. https://doi.org/10.5465/amj.2005.15993166  

Public Company Accounting Oversight Board. (2023). Staff report on audit quality in connection with the use of technology tools. https://pcaobus.org  

van Dijk, L., Frick, J., Thurkow, T., Mottura, L., & Hubert, A. (2026, May). An operating model for the age of AI. Bain & Company. https://www.bain.com/insights/an-operating-model-for-the-age-of-ai/  

Craig Friedman Bio 

Craig Friedman has 30 years of experience as a human capital and talent advisor, executive, and entrepreneur. He specializes in skills-based talent strategies, learning operating models, change management, and performance consulting. A Senior Talent Strategist at St. Charles Consulting Group, Craig partners with CHROs and CLOs to align global talent strategies, supporting Fortune 500 companies and four of the five largest professional services firms. 

Previously, Craig spent 15 years at Deloitte, leading talent development for the U.S. Tax practice and earning several national and international learning awards. He also co-led Deloitte’s clinician change adoption practice and helped establish corporate universities for regional health systems. 

Craig’s background includes launching two eLearning start-ups and earning two U.S. patents for innovations in online education. He holds an M.A. in Learning Sciences from Northwestern University and dual undergraduate degrees from Tufts University in Human Factors Engineering and English. Craig is the author of the new book Enterprise Skills Unlocked: A Blueprint for Building a Skills-Based Organization, published in June 2025.