The traditional public accounting model has long been a pyramid structure: firms hire large classes of staff associates, hand them the high-volume, low-judgment work, and select a few standouts to advance each year. The entry-level work wasn’t just billable; it was an apprenticeship. Up-and-comers learned what a clean confirmation looked like by reviewing what surely felt like a thousand of them. Artificial intelligence (AI) is dismantling the base of that pyramid, and the profession hasn’t fully reckoned with what will happen to the entire structure.
Meanwhile, firms are advertising for prompt engineers, agent-automation managers, and people who can help clients “embrace” generative AI. The pyramid isn’t just getting narrower at the bottom; its composition is changing altogether.
Is it any wonder? Firms have invested significant capital in AI. PwC committed $1 billion and became one of OpenAI’s largest enterprise customers. KPMG struck a multibillion-dollar alliance with Microsoft. Deloitte and EY each invested over a billion developing proprietary platforms, and Deloitte deployed Claude to its global workforce. EY has reported putting 150-plus AI agents in front of tens of thousands of tax staff. These aren’t pilots anymore.3
Outside the Big 4, as private equity continues to invest in accounting firms, AI is being viewed as one of the primary value-creation levers. Private equity capital is being deployed to accelerate AI adoption, automate labor-intensive work, and create new advisory revenue streams.
It helps to be concrete because “AI is transforming audit” is a very broad sentence that lacks much meaning. A staff associate’s day is full of tasks that generative and agentic tools genuinely do well, such as document review (pulling terms out of leases and revenue contracts; flagging the clauses that matter for ASC 842 or 606), financial statement tie-outs and reconciliations, and anomaly and fraud pattern detection, to name a few.
None of this eliminates the need for professionals on engagements. A human still signs the opinion, owns the client relationship, and carries the regulatory risk. Leveraging AI really just compresses the hours, and those hours were historically mostly staff associate hours. When a tool drafts a memo and tests the full population in minutes, the firm doesn’t need five first-year associates on that workstream. It needs one (two tops) who can supervise the output.
Here’s the part that not a lot of people think about: the entry-level work that AI is absorbing was never only about production; it was how staff learned judgment. Accountants developed a sense for what’s reasonable – what a normal margin looks like, when a reconciling item doesn’t pass the sniff test, account variance expectations – by doing the repetitive work, getting review comments, and doing it better. Remove the bottom of the pyramid and you damage the mechanism that produces seniors, managers, and eventually partners.
Firm leaders are increasingly candid that the skills they want at the entry level are now different. Surveys of finance and accounting leaders consistently rank strategic thinking and problem-solving above technical execution. The framing you hear is that juniors are becoming “orchestrators” or “architects” designing and reviewing automated processes rather than performing them.4 That’s a real and useful reframing; it’s also a much harder job to entrust to a new hire fresh out of college who has never seen the underlying work done by hand. You can’t provide meaningful review of an AI-drafted lease memo if you’ve never drafted one yourself and don’t know where a tool may tend to be wrong.
Work that resists automation is the work that was always the point: professional skepticism, client judgment, the ability to ask whether the convincing-looking output is actually (materially) correct. Generative AI outputs are fluent and confident … and routinely wrong in ways that matter. A hallucinated authoritative citation, a plausible-but-incorrect technical conclusion, a number that foots but shouldn’t exist are real pitfalls. The associate who can catch these becomes more valuable, not less. At Centri’s 2025 Capital Conference,5 the panel Cutting Through the Hype: How Can Growth-Stage Investors and Company CEOs Make the Most of AI opened with a succinct message: AI won’t replace you, but someone using [it] might.6
The entry-level profile is genuinely shifting toward a hybrid: solid technical accounting fundamentals, the AI fluency to drive the tools, and the skepticism to distrust them. For staff associates and the firms training them, that suggests a few practical priorities: learn the underlying accounting deeply enough to audit the machine and not just push the prompts; get fluent with the firm’s actual platforms rather than treating AI as a side skill; and invest early in the human-facing capabilities (client communication, judgment, and the ability to explain why) that the tools can’t replicate and that the firms can no longer assume associates will learn by doing.
Small and midtier firms face a sharper version of this than the Big Four. These firms can’t write billion-dollar checks for proprietary platforms, but they also can’t afford to keep staffing their engagements in the old way while competitors automate. Their path is to buy and integrate tools effectively while competing on the judgment and relationships that scale poorly. The most practical solution is a dual-track model: automate the lowest value, repetitive tasks, but deliberately reintroduce “synthetic reps” elsewhere so staff still accumulate the judgment those tasks once developed. That can take several forms: AI-with-explanation workflows, where automation outputs are not simply accepted but must be reviewed, annotated, and explained by staff; tiered review layers that emphasize coaching over correction; or building advisory-ready skill paths earlier in careers with emphasis on AI supervision specializations.
Diving deeper, these hard questions are even being considered at the university level with programs asking, “If AI eliminates much of the traditional staff-level work, how do you train the future accountants?” According to reporting on programs at institutions such as Georgia State, Notre Dame, and Brigham Young University, among others, schools are incorporating AI directly into accounting coursework rather than treating it as a separate technology class. Students are learning how to use AI tools in audit and risk assessment scenarios.7
The optimistic reading is that AI elevates the work: fewer hours ticking and more time on analysis and advisory. There will be a faster progression for the people who adapt. A pessimistic reading would see a hollowed-out talent pipeline: firms that automated away the training ground and discover, five years out, that they didn’t develop enough people who can supervise the automation. Two things can be true at once, of course. The firms that thrive will be the ones that figure out how to develop judgment when the work that used to build it has been handed to a machine.
1 “The Big Four’s New Favourite Grad is AI,” Accountancy Age (June 2025).
2 Polly Thompson, “Getting a Job at PwC Out of College Will Be a Lot Tougher. It Plans to Recruit a Third Fewer Grads by 2028,” Business Insider (August 2025).
3 Marin Ivezic, “2026 Consulting’s AI Revolution Update,” FutureOfConsulting.AI (January 2026).
4 Demi Lawrence and CFO Brew, “Accounting’s Big ‘Wake-Up Call,” Fortune (May 2026).
5 https://centriconsulting.com/capital-conference
6 https://centriconsulting.com/news/insights/how-growth-stage-leaders-can-harness-ai-for-real-impact-5-key-takeaways
7 https://news.bloombergtax.com/financial-accounting/colleges-inject-ai-into-accounting-programs-to-increase-appeal
Rikki Williams, CPA, is a senior director within Centri Business Consulting’s National Office. He serves as the firm’s subject matter expert for complex financial instruments. He can be reached at rwilliams@centriconsulting.com.
Michael Kirchner is a managing director at Centri Business Consulting. He can be reached at mkirchner@centriconsulting.com.