Anthropic builds Claude, the AI that now reads receipts, codes transactions and drafts month-end reports inside accounting firms. Its own finance team uses that AI more than most, and it still hires human accountants, and a person still signs off on the numbers before they are filed. If even the company building the technology cannot take the human out of its own books, the useful question is not whether AI replaces accountants. It is which parts of the job it takes, and which parts it hands back.
The Bureau of Labor Statistics projects 5 percent growth in accountant and auditor employment over the 2024 to 2034 decade, faster than the average for all occupations. Over the same period, it projects a 6 percent decline in bookkeeping, accounting, and auditing clerk positions. Goldman Sachs, in a separate analysis, rated accountants and auditors at the highest risk of AI displacement based on task exposure. All three findings are accurate, and they point in the same direction once you separate what the data is actually measuring. This article works through 21 data points from government statistics, academic field studies, and professional body surveys to explain why the profession is contracting in some places and growing in others. For more background statistics, the 75-stat roundup covers the broader AI-in-accounting picture.
Are accountants being replaced by AI?
The Goldman Sachs finding and the BLS projection look contradictory until you see they are measuring different things. Goldman Sachs analyzed task exposure: what percentage of an accountant's daily tasks could theoretically be performed by AI given current technology. The BLS projection measures expected employment headcount across the decade. Task exposure tells you which tasks a role contains that AI could potentially handle. Employment projection tells you whether firms will employ more or fewer people in that role overall. A role can have high task exposure and growing employment at the same time, if the value of the remaining human work increases as the automatable tasks are handled by software.
The Bureau of Labor Statistics projects 5 percent growth in accountant and auditor employment between 2024 and 2034, faster than the average for all occupations (BLS Occupational Outlook Handbook). Over the same decade, it projects a 6 percent decline in bookkeeping, accounting, and auditing clerk positions (BLS). Meanwhile, Goldman Sachs rates accountants and auditors as facing the highest risk of displacement by AI based on task exposure analysis, not employment projections (Goldman Sachs Global Economics Research). The distinction matters: task exposure measures theoretical automation potential, not actual job loss. Professional roles grow while clerical roles shrink, and task exposure is the thread connecting both trends.
Both BLS trends run along the same dividing line: tasks AI handles well today versus tasks it does not. Clerks spend most of their time on data entry, transaction coding, invoice matching, and document processing. That work is high-volume, pattern-based, and structurally well suited to machine learning. Accountants spend more of their time on interpretation and professional judgment, work that depends on client context. AI is absorbing the tasks in one role and leaving the other intact.
Does the company that builds the AI still hire accountants?
Anthropic runs one of the most automated finance functions around, and its own figures show where the line currently sits. CFO Krishna Rao has said the team's monthly financial review is 90 to 95 percent ready before a person steps in, and that reporting work which once took hours now takes about 30 minutes (Business Today). The AI gathers the data and produces the first draft. People decide what it means. Rao describes the shift as moving staff away from collecting information and toward interpreting it, so that "everyone kind of becomes a manager".
That last stretch is the part the technology does not close on its own. Anthropic's own finance product is built to keep users "firmly in the loop," reviewing, iterating on, and approving the work before anything is filed (Anthropic), and the company is still recruiting in-house accountants and finance staff to do it. It is the whole pattern in one company: the high-volume work compresses, the judgment work does not, and the people stay for the second kind. That is the same split described in what AI can and cannot do in a practice, here visible inside the firm that makes the tools.
Why did 340,000 accountants leave?
The departure started before AI tools were production-ready for accounting work. People left because of burnout, a retirement cliff that has been building for two decades, and a pipeline that dried up as fewer students chose the profession. The people who stayed did so through a period of mounting pressure that has not fully resolved.
340,000 accountants left the US profession between 2019 and 2023, a 17 percent decline from approximately 1.6 million. The figure comes from the largest workforce study the profession has conducted: approximately 8,000 survey responses and briefings with 15,000 participants. (AICPA/NASBA/NPAG Accounting Talent Strategy Report)
Accounting graduates hit a 20-year low in 2023 to 2024: 55,152 graduates, down 6.6 percent year on year. Master's degrees in accounting fell 15 percent over the same period. (AICPA/CIMA 2025 Trends Report)
Nearly 75 percent of CPAs in the United States are at or near retirement age, according to AICPA pipeline data. The credential figure is specific to the US credentialing system. (AICPA pipeline data, reported by Ramp)
86 percent of accountants report burnout, and 25 percent say they are seriously considering leaving the profession within the next year. (Sage "Practice of Now" survey, n=1,000 across 6 markets)
None of those four numbers have anything to do with AI. Firms that lost staff to burnout in 2021 and 2022 lost them to overwork. The workforce is now smaller and older, and the same volume of compliance work still needs doing. AI tools are entering that environment, which means they are filling a gap rather than displacing people from stable jobs.
What is happening to bookkeeping and clerical roles?
Roles the BLS expects to decline are built around tasks that fit AI's current strengths: data entry, transaction categorization, invoice matching, bank reconciliation. High-volume work on structured data with predictable rules. Roles the BLS expects to grow depend on interpretation, advisory work, and professional judgment, which AI still handles poorly. What matters is whether automatable tasks make up most of a given job or just a fraction of it.
McKinsey Global Institute puts two numbers on this shift: 42 percent of finance and accounting tasks can now be automated using current technology (McKinsey), and 27 percent of all current work hours are projected to be automated by 2030, with accounting identified as facing the greatest disruption among white-collar professions (McKinsey). The gap between the 42 percent that is technically automatable and the 27 percent projected to actually be automated reflects how slowly firms adopt even when the technology exists.
Manual invoice processing costs $15.97 per invoice; automated processing brings that to $3.24, roughly an 80 percent reduction in per-invoice cost. (IOFM / Ardent Partners, reported by Precoro)
That invoice figure is worth sitting with. When the most time-consuming task in a job can be done for $3.24 instead of $15.97, the case for hiring a person to do only that task weakens considerably. The roles that hold up are the ones where a person handles what the software cannot: exceptions, client contact, professional review, and calls that require context the system does not have.
How is AI changing the work that remains?
The Stanford GSB and MIT Sloan field study published in 2025 by researchers Choi and Xie is the most useful data on this question. It tracked 277 accountants across 79 firms and observed actual outcomes, rather than asking respondents what they believed or intended. Field studies of this kind are harder to run than surveys, and the sample is smaller, but the findings are grounded in what happened in practice.
The field study found that AI-using accountants cut 7.5 days off their monthly close and supported 55 percent more clients per week compared to peers not using AI tools (Stanford GSB / MIT Sloan, Choi and Xie 2025, n=277 accountants + 79 firms). Perhaps more telling for the shape of the profession, 8.5 percent of working time shifted from data entry to advisory and review work, approximately 3.5 hours per week moved from manual processing to client-facing tasks (Stanford GSB / MIT Sloan).
Cutting 7.5 days from the monthly close saves time. Shifting 3.5 hours a week from data entry to advisory work changes what the job actually involves. That second number is the one that reshapes the profession. The full study data is in the 75-stat roundup.
Advisory now accounts for 13 percent of firm revenue, up from 10 percent in 2024. 93 percent of firms now offer some form of advisory service. (Wolters Kluwer Future Ready Accountant Report 2025, n=2,768 across 14 countries)
Three percentage points sounds modest. But advisory was barely trackable as a revenue category in most small firms three years ago, and the Wolters Kluwer data suggests it is now a tenth of firm income across 14 countries. Compliance work takes less time when AI handles routine steps; the hours freed up have to go somewhere, and advisory is where they are going.
Who is adopting AI fastest?
The generational data here goes against the obvious prediction. Younger accountants use AI more often. Older practitioners use it for client work where their experience helps them spot when an AI output misses something. Frequency and quality of use are not the same thing.
The Chartered Accountants Worldwide and Ipsos survey (n=2,718 across 48 countries) found that 83 percent of chartered accountants aged 18 to 24 use AI at least weekly, and 80 percent of that age group feel confident using AI tools, compared to just 55 percent of senior decision-makers. (Chartered Accountants Worldwide / Ipsos) The confidence gap is wide, but it does not tell the full story about who generates value from AI.
One UK survey found the opposite pattern when looking specifically at client work: 50 percent of accountants aged 55 and older use AI for client insights, compared to 16 percent of those aged 18 to 24. (AccountingWEB UK) This is a single survey, smaller and less rigorous than the Chartered Accountants Worldwide data, so treat the number as a signal. But the direction makes sense. Senior practitioners have the client relationships to apply AI to, and the professional experience to know when an AI output is right, wrong, or incomplete. Junior staff use AI more often, mostly for research and drafting, not for the conversations that generate revenue.
For firms thinking about where to direct training budget, this argues against concentrating it entirely on junior staff. A senior practitioner already using AI for client work probably generates more value per hour of training than a junior employee learning general AI literacy. Specific tool training for people who already have context beats broad courses for people who do not.
What is stopping firms from adapting?
The perception gap in this data is wide. Finance professionals broadly agree that AI is consequential. Very few feel ready for it.
The AICPA and CIMA Future-Ready Finance Survey (n=1,446) found that 88 percent of finance professionals believe AI will be the most transformative technology they face, but only 8 percent feel "very well prepared" to use it. In the same survey, 56 percent identify generative AI as the most prominent skills gap in the profession today. (AICPA and CIMA) Meanwhile, only 37 percent of accounting firms invest in AI training, according to a Karbon survey of accounting professionals (Karbon State of AI in Accounting 2025. Note: vendor survey of its own user base, so it may skew toward firms already engaged with AI).
The gap between the 88 percent who believe AI is transformative and the 37 percent investing in training is where the friction is. A firm that has seen the headline numbers and decided AI matters, but has not yet assigned time or budget to act on it, is running a risk. Tools are getting more capable. Firms that invested early will show measurably different output per person within the next two or three years. The ones watching from the sidelines will notice it when they lose staff or clients to firms that did.
Some evidence suggests the pipeline may be turning. Accounting enrollments grew 12 percent year on year for two consecutive semesters in 2024 to 2025, the first sustained reversal after years of decline. (National Student Clearinghouse, reported by Journal of Accountancy) Whether that represents a real shift in how the profession is perceived by prospective students, or a temporary fluctuation, will take several more years of data to determine. The direction is at least better than it was.
Twenty-one data points, and they point the same way: the profession is not disappearing, but its shape is changing faster than at any point in recent memory. Roles built around transactional tasks are contracting. Roles built around professional judgment and client advisory are growing. The firms that come through it are the ones that have already started closing the training gap. For the full picture across 27 primary sources, the 75-stat roundup covers market size, adoption rates, time savings, and buyer behavior.
