Will AI replace the people who keep the books?
The U.S. Bureau of Labor Statistics projects bookkeeping clerks down 6 percent by 2035, accountants up 5 and executive assistants flat. Accounting benchmarks from 2025 and 2026 show agents far short of a full close. Read on 2026-11-05.
Contents
Short answer, from the sources below: the official projections do not say AI replaces accountants or executive assistants by 2035, they do say fewer bookkeeping clerks will be needed, and they blame software that already exists. The first measurable effect of AI on jobs is on hiring young workers in the occupations most exposed to it. This page lists what each source says, who wrote it and what it leaves out.
What the projections say
The U.S. Bureau of Labor Statistics (BLS) publishes ten-year employment projections for each occupation. The 2025-35 round was read on 2026-11-05; BLS shows total employment across all occupations growing 3 percent.[1]
- Bookkeeping, accounting and auditing clerks: 1,532,400 jobs in 2025 to 1,446,800 in 2035, a fall of 6 percent.[1] BLS gives its reason: "Technological change is expected to reduce demand for these workers. Software innovations have automated many of the tasks performed by bookkeeping, accounting, and auditing clerks."[1] The reason names software, not AI. BLS still expects about 144,100 openings a year, all from workers who leave or retire, and a median wage of $50,670 in May 2025.[1]
- Accountants and auditors: 1,595,200 jobs in 2025 to 1,674,600 in 2035, up 5 percent.[2] BLS does name AI here: "Although these technologies will likely increase accountants' efficiency, this change is not expected to reduce overall demand."[2] The median wage was $83,680 in May 2025.[2]
- Secretaries and administrative assistants: 3,515,600 jobs to 3,440,300, down 2 percent overall.[3] Within that, executive secretaries and executive administrative assistants go from 489,300 to 489,200, a change of -100 or 0 percent.[3] Legal fall 5 percent, medical rise 5 percent, and the remaining group falls 6 percent.[3] BLS names AI for the groups that fall: "Technology, including artificial intelligence (AI) systems and digital tools, enables staff in many organizations to prepare their own documents without the help of secretaries."[3] It adds that many managers "now do tasks that were previously done by these workers."[3]
For the occupations that shrink, BLS expects the people who remain to do different work. For clerks it says they "are expected to take on a more analytical and advisory role over the decade", and for accountants that routine tasks such as data entry "will instead make accountants' advisory and analytical duties more prominent."[1][2]
A projection is an estimate made from past trends; it is not a count of what AI did. These BLS pages are the agency's own summary text; its method notes are a separate document.
Canada
Canada's Job Bank gives its own rating for bookkeepers. The projections from the Canadian Occupational Projections System for 2024-2033 say "Labour demand and labour supply are expected to be broadly in line for this occupation over the period of 2024-2033 at the national level."[4] There were 204,400 bookkeepers in 2023, 51 percent were aged 50 or over, and the median retirement age was 65.0.[4] Job Bank rates the job prospects over the next 3 years Moderate in British Columbia, Manitoba and Yukon, Very good in Quebec and the Northwest Territories, and Good in the other provinces and territories, including Alberta and Ontario.[4] The page says nothing about AI. Job Bank outlooks for accountants and assistants are separate pages.
What agents can do on the work itself
Two benchmarks test agents on bookkeeping and accounting work. Both were built by companies (Penrose; Mercor with Ramp), not by a neutral body, so read them as the builders' own measurements.
- AccountingBench (Penrose, 2025): agents get one year of data from a real software business and must "close the books" month after month, with a CPA's work as the baseline.[5] Penrose reports that "Grok 4 and Claude 4 tend to perform well initially (within 1% of CPA baselines), but accumulate material errors over time", and that "O3, O4-Mini and 2.5 Pro were unable to close 1 month of books, giving up partway through."[5] Grok 4 and Claude 4 are models of mid-2025.
- APEX-Accounting (Mercor with Ramp, July 2026): 160 tasks in 10 simulated companies, written and solved by experts in accounting and bookkeeping, covering reconciling accounts, accruing expenses and posting transactions.[6] Across nine frontier models the leader scores 56.4 percent on the paper's mean-criteria measure scored against the experts' rubrics, "No model scores more than 2.6% Pass^8" and the highest Pass@8 is 21.5 percent.[6] Pass^8 is the pass^k measure that τ-bench introduced "to evaluate the reliability of agent behavior over multiple trials", read here as the share of tasks an agent gets right on all eight tries.[8] The measures are explained, with other tests, in Do AI agents actually work?.
Read together: on the same kind of work, the 2026 results are far from a full replacement, and the 2025 results found that early accuracy did not hold over months. Neither benchmark measures a bookkeeper using a model as a tool and checking its output, which is how most firms use it.
Who is affected first
Stanford's Digital Economy Lab used payroll data from ADP covering millions of U.S. workers through June 2026. In the revised paper (12 August 2026) the authors "find no evidence of widespread, economy-wide job displacement", and also that "employment of young workers (ages 22-25) in AI-exposed occupations now stands 19% below where it would be had it kept pace with that of their less-exposed peers; experienced workers show no comparable gap."[7] The gap "operates primarily through reduced hiring of young workers rather than increased separations."[7] The authors call these "early, descriptive indicators" and not causal estimates, and say the pattern weakens when they control for education.[7] The abstract does not name bookkeeping or accounting, so the size of the effect for them is not stated.
Not yet proven
- Whether the projections will hold. BLS revises its projections every year, and the 2026 benchmark results are newer than most of the data behind them.
- Canadian accountants and assistants. Job Bank's outlook is cited here for bookkeepers only.
- Exposure scores. Studies that rate how much of each job a model could do are not compared here.
- Firm-level evidence. No survey of firms that cut bookkeeping staff because of AI is cited here.
If you run a bookkeeping practice and have seen AI change who you hire or what you charge, I would like to compare notes: mackenzie@mackenziebowes.com.
Sources
- U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Bookkeeping, Accounting, and Auditing Clerks. Last modified 2026-08-27, read 2026-11-05. www.
bls. gov/ ooh/ office- and- administrative- support/ bookkeeping- accounting- and- auditing- clerks. htm - U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Accountants and Auditors. Last modified 2026-10-29, read 2026-11-05. www.
bls. gov/ ooh/ business- and- financial/ accountants- and- auditors. htm - U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Secretaries and Administrative Assistants. Last modified 2026-08-27, read 2026-11-05. www.
bls. gov/ ooh/ office- and- administrative- support/ secretaries- and- administrative- assistants. htm - Job Bank (Government of Canada), Bookkeeper in Canada, job prospects. Page modified 2026-09-24; outlooks updated 2025-12-10; projections from the Canadian Occupational Projections System (ESDC). www.
jobbank. gc. ca/ marketreport/ outlook- occupation/ 12562/ ca - Penrose, AccountingBench: Evaluating LLMs on Real Long-Horizon Business Tasks (2025; the page carries no date). accounting.
penrose. com/ - Benchek et al. (Mercor), APEX-Accounting, arXiv 2607.27189, submitted 2026-07-29, revised 2026-07-30 (abstract page). arxiv.
org/ abs/ 2607. 27189 - Brynjolfsson, Chandar and Chen (Stanford Digital Economy Lab), Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, working paper revised 2026-08-12 (publication page with abstract). digitaleconomy.
stanford. edu/ publication/ canaries- in- the- coal- mine- six- facts- about- the- recent- employment- effects- of- artificial- intelligence/ - Yao, Shinn, Razavi and Narasimhan, τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains, arXiv 2406.12045, submitted 2024-06-17 (abstract page). arxiv.
org/ abs/ 2406. 12045
Terms on this page
- AI agent: A system where a language model decides for itself how to do a task and which tools to use.