Will AI Replace Accountants?

What the employment figures say

The US Bureau of Labor Statistics counts 1,579,800 jobs in the Accountants and Auditors category, SOC 13-2011, in 2024, and projects 5% growth to 2034, an increase of 72,800 ^{(US Bureau of Labor Statistics 2025)}. That does not agree with a Critically Exposed score, and the disagreement is real rather than an error: the category bundles accountants with auditors and spans a broad range of financial-record preparation and examination work, and the authority does not attribute the projected growth to AI ^{(US Bureau of Labor Statistics 2025)}. Roughly 124,200 openings a year are expected on average, many of them from workers transferring occupations or leaving the labor force, including retirement ^{(US Bureau of Labor Statistics 2025)}. Exposure describes the task content; the headcount reflects how slowly regulated organisations change shape.

What the research says about accountants

Now

Accountants and Auditors are marked fully exposed by Eloundou, high on Felten's AIOE, and listed by employers among the fastest-declining roles — an unusual level of agreement across three independent measures ^{(World Economic Forum 2025)}. Adoption is real but partial: 21% of tax firms already use generative AI and 53% plan to or are considering it, with 25% reporting no current plans, down from 49% in 2024 — figures from Thomson Reuters, which sells software to the market it surveyed ^{(Thomson Reuters Tax & Accounting 2025)}. Where the tools land, they compress time rather than headcount: a customised memo bot at the practice Hopper cut an accounting memo from four hours to 30 minutes with human review retained ^{(Journal of Accountancy 2025)}, and a Stanford field study of 277 accountants at 79 firms — run with an AI accounting-software partner — found complementarity, with a 55% rise in weekly client support and monthly close 7.5 days faster ^{(Stanford Graduate School of Business 2025)}. Across the same stretch the official series shows employment rising from 1,280,700 in May 2019 to 1,579,800 in 2024, and median pay from $71,550 in 2019 to $81,680 in May 2024 ^{(US Bureau of Labor Statistics 2024) (US Bureau of Labor Statistics 2025)}.

The next few years

Deloitte's audit platform applies generative and agentic capabilities to reviewing audit documentation, EY technology supports more than 160,000 global audit engagements, and PwC reports 20%-50% productivity gains in development work while planning an end-to-end AI-driven audit solution for 2026 — as reported by Thomson Reuters, a vendor with a commercial interest in adoption narratives ^{(Thomson Reuters Tax & Accounting 2025)}. Agentic systems shift the question from whether AI can help with a task to whether it can finish the task unattended, and routine reporting and data enrichment move first into scheduled workflows while interpretation and decision ownership lag ^{(Anthropic Economic Index 2026) (Microsoft Research 2025)}. Pulling the other way is skills supply: ACCA's 2025 survey of more than 10,000 professionals in 175 countries found only 32% said their organisation provides opportunities to learn AI-related skills, and half were concerned they were not developing the skills the future workplace requires ^{(ACCA 2025)}. Firms are still recruiting — 11,985 new graduate hires in 2024, with 75% of hiring firms expecting similar or higher hiring in 2025 and 18% expecting lower — though AICPA has an institutional interest in recruiting into the profession and does not attribute these movements to AI ^{(AICPA & CIMA 2025)}.

Longer term

The longer-term floor here is regulatory rather than technical. PCAOB AS 1000 requires the auditor to exercise professional judgment, applying training, knowledge and experience to make informed decisions, which keeps a named person accountable even when technology performs the analysis ^{(PCAOB n.d.)}. The board's 2024 standards action on technology-assisted analysis, approved by the SEC on 20 August 2024, clarified auditor responsibilities when such analysis is used rather than transferring accountability to a tool ^{(PCAOB 2024)}. O*NET frames the occupation around examining, analysing and interpreting accounting records to prepare statements, give advice or audit others' statements, leaving skeptical evaluation, final review and regulated audit responsibility as the least automatable layer ^{(O*NET OnLine n.d.)}.

What is exposed, and what protects it

● Where the exposure is

The exposed core is quantitative and document work: ledger maintenance, reconciliation and routine reporting, which is where the exposure measures concentrate and why three of them agree ^{(World Economic Forum 2025)}. Drafting and analysis compress sharply in practitioner accounts — hours to minutes, with review retained, and roughly 60 hours of setup to get there ^{(Journal of Accountancy 2025)}. Appetite is not the constraint either: an ICAEW and Chartered Accountants Worldwide survey of 2,718 respondents across 48 countries found 85% willing to use AI if given the opportunity, though those bodies represent and develop the profession and the survey measures perceptions rather than labour-market outcomes ^{(ICAEW and Chartered Accountants Worldwide 2025)}.

○ What holds

The moat is regulation and licensure, not anything physical. Audit standards require professional judgment from a person, and the 2024 clarification placed responsibility for technology-assisted analysis on the auditor rather than the software ^{(PCAOB n.d.) (PCAOB 2024)}. Deployment therefore lags capability inside regulated work, which is part of why the official employment series has grown rather than shrunk ^{(US Bureau of Labor Statistics 2024)}. Practitioner accounts show the limits close up too: a client-facing advisory tool built with ChatGPT at a 15-person firm was described as about 95% factual but weak on calculations, with staff told to double-check answers ^{(Journal of Accountancy 2025)}.

Which tasks go first

Already covered: transaction coding, data entry, first-draft memos and standard report production — the Stanford field data shows about 8.5% of time reallocated away from routine data entry and a 12% rise in general-ledger granularity as software handles more of the detail ^{(Stanford Graduate School of Business 2025)}. The agentic shift moves routine reporting and data enrichment out of prompted assistance and into API workflows that run multi-step without a person at each step, while framing and decision ownership stay behind ^{(Anthropic Economic Index 2026) (Microsoft Research 2025)}. Inside large firms the same pattern appears in audit-documentation review, which is where generative and agentic capability has been pointed first ^{(Thomson Reuters Tax & Accounting 2025)}. What stays human is the judgment layer O*NET describes: interpreting records, skeptical evaluation, final review and the signature on regulated work ^{(O*NET OnLine n.d.) (PCAOB n.d.)}. In the field study the freed hours did not vanish — they moved toward client support and quality assurance while close times fell ^{(Stanford Graduate School of Business 2025)}.

What to do about it

Move toward the work that carries a signature

The protected layer is regulated judgment: professional scepticism, final review and named accountability for technology-assisted analysis ^{(PCAOB n.d.) (PCAOB 2024)}. Time spent on reconciliation and report production is the part the exposure measures target ^{(World Economic Forum 2025)}. Shifting your task mix toward review, advisory and sign-off work moves you into the layer that regulation currently reserves for a person.

Do not wait for an employer training programme

Only 32% of accountancy and finance professionals said their organisation offers opportunities to learn AI-related skills, while half were concerned about their future skills ^{(ACCA 2025)}. Meanwhile 52% of tax-firm users are working with open-source tools such as ChatGPT rather than industry-specific software ^{(Thomson Reuters Tax & Accounting 2025)}. That gap is one an individual can close on their own with the same general tools the early adopters are using.

Build the client-facing half of the role

Where AI landed in the field study, weekly client support rose 55% and time moved toward communication and quality assurance ^{(Stanford Graduate School of Business 2025)}. Practitioner accounts point the same way, with tools used to build advisory material that staff then check ^{(Journal of Accountancy 2025) (EY US 2024)}. Relationship and interpretation work is where the freed capacity is going, which makes it the sensible place to be strong.

Tools and courses that fit this situation

Below are paid tools and courses that fit where this occupation sits: general AI skills training, payroll and compliance software, and a resume service if your next move is out of a transactional role. None of them change the exposure of the underlying tasks.

Some links above are affiliate links. If you buy through them, Jobs After AI may earn a commission at no extra cost to you. It does not change which tools appear here.

Common questions

Is this about accountants losing jobs, or about the tasks changing?

It is about tasks. The exposure measures describe how much of the job's content current AI can do, and the official employment series has risen from 1,280,700 in May 2019 to 1,579,800 in 2024 with median pay also rising ^{(US Bureau of Labor Statistics 2024) (US Bureau of Labor Statistics 2025)}. Field evidence so far points to time being reallocated rather than headcount removed ^{(Stanford Graduate School of Business 2025)}.

How is the score of 32 out of 100 worked out?

It is a composite of task-exposure research, and accountants sit low because three independent measures agree: fully exposed under Eloundou, high on Felten's AIOE, and among employer-reported fastest-declining roles ^{(World Economic Forum 2025)}. Confidence here is medium, so read 32 as directional rather than precise. It measures task exposure over a three-to-five-year view, not a forecast of employment.

Are junior accountants affected first?

That is the pattern to watch. There is no systematic rise in unemployment among exposed workers, but the clearest early signal is a tentative slowing of hiring for under-25s in exposed roles ^{(Anthropic Economic Index 2026)}. In accounting the entry pipeline has thinned — new CPA Exam candidates fell from 42,626 in 2023 to 28,082 in 2024 — though AICPA attributes that to entry-level pay, costs and administrative hurdles rather than AI ^{(AICPA & CIMA 2025)}.

What part of this job is AI still bad at?

Calculation reliability and accountability. A practitioner-built advisory tool was described as about 95% factual but struggled with calculations, with staff instructed to double-check ^{(Journal of Accountancy 2025)}. Audit standards still require professional judgment from the auditor, and the 2024 clarification kept responsibility with the person when technology performs the analysis ^{(PCAOB n.d.) (PCAOB 2024)}.

Should I still train as an accountant?

The official projection is 5% growth to 2034 with roughly 124,200 openings a year on average, many from transfers and retirements, and the authority does not attribute that to AI ^{(US Bureau of Labor Statistics 2025)}. Enrollment grew 12% year over year in two consecutive semesters in 2024-25 even as graduate numbers fell to 55,152 ^{(AICPA & CIMA 2025)}. The route is open; the task mix you build toward matters more than it used to.

Related occupations

The evidence behind this page

How this score was built

The AI-Proof Score is an estimate of how exposed an occupation's core tasks are to AI over roughly the next three to five years, on a 0–100 scale where higher is safer. It measures the exposure of tasks, which sits upstream of, and is softer than, job loss — it is not a probability that this job disappears, and not a claim about any individual doing it. The full derivation, the four research families behind it and its limitations are set out in the methodology, and every occupation is ranked in the AI Job Impact Report 2026.