AI in Accounting: What's Actually Changing by Job Function

The profession is splitting into two tracks. The Bureau of Labor Statistics projects accountant and auditor employment growing 5% through 2034 — 124,200 openings per year — while simultaneously projecting payroll and timekeeping clerk roles declining 16.7% over the same period. That's not a contradiction. That's the story.

Stanford researchers Choi and Xie studied 277 accountants across 79 firms and found AI closes monthly books 7.5 days faster and cuts back-office time 8.5%. But the gains went overwhelmingly to senior accountants who verified AI outputs — not junior staff who accepted them. The tool rewards judgment. If you have it, AI amplifies it. If you don't yet, AI exposes the gap.

Data-entry bookkeeping and payroll-clerk roles face real near-term pressure. Audit, tax strategy, and advisory roles are growing and transforming. Across every function, the accountants doing best treat AI as a collaborator to verify, not an oracle to trust. Here's what that looks like depending on where you sit.

If You Work in Bookkeeping, AP, or Payroll

This is where the real displacement pressure is. Xero's data puts 80-90% of routine bookkeeping tasks — transaction classification, bank reconciliation, expense categorization — as automatable with current tools. That's not a 2030 projection. Vic.ai (which has raised $63M specifically for autonomous invoice processing), Stampli, and Ramp are doing this today for clients. The bookkeeping profession has already contracted 5% as AI adoption grew. If your primary value is moving numbers from receipts into a spreadsheet, that task is under genuine near-term pressure.

What's not under pressure is what Karl Spanbauer, CPA, controller at Capital Area Food Bank, represents. Spanbauer faced hundreds of pieces of mail per week — invoices, tax notices, grant letters — manually sorted and routed. "I lost a couple documents. It was literally hundreds of pieces of mail," he told the Journal of Accountancy. His solution: scan everything, run it through a secure LLM to extract and summarize, auto-route a Jira ticket to the right person via Microsoft Power Automate. Result — 4 hours a week recovered, compressed into a 20-minute "Mail Mondays" session.

His most important observation: "AI can't interact with a piece of paper sitting on my desk. I had to get that piece of mail into that ecosystem." Digitization is the prerequisite. The automation follows.

AI can't interact with a piece of paper sitting on my desk. I had to get that piece of mail into that ecosystem.
— Karl Spanbauer, Controller, Capital Area Food Bank

That's what non-displaceable bookkeeping looks like — not entering data, but architecting the flow of data. The bookkeeper who learns to configure and supervise tools like Vic.ai or FloQast (the leading platform for month-end close at mid-market companies) is not replaced by those tools. They become the person who runs them. If your work is 80%+ data entry right now, this section is urgent.

If You Work in Audit

Audit is a different story. The exposure is real, but the direction is different — AI is changing procedures faster than almost any other function, while the thing AI can't do remains exactly what makes auditors valuable.

Natalie Denman, CIA, audit supervisor at Flowserve Corporation, uses Workiva's generative AI for three-way match testing on purchase orders, converting walkthrough transcripts into editable documents, and sorting executive survey responses for risk assessments. Real efficiency gains, real time savings. Then a chatbot produced an unrealistic revenue figure, she used it in a stakeholder communication, and the blowback was immediate. "You can't ever blindly trust this tool." She's a strong AI advocate — which makes the warning more credible, not less.

The lesson isn't don't use AI. It's that verification is now a core audit competency.

Tricia Katebini, CPA, MBA, partner at GRF CPAs & Advisors, is more uniformly positive. Trullion's Audit Suite turned scouring through 80-page bulk PDFs — board minutes, lease agreements, loan covenants — into something entirely different. "You're not scouring through 80 pages of a bulk PDF. It's very easy in that respect." Footings and disclosure tying are now automated: "Our production team doesn't have to go through the process of manually footing everything... It's the press of a button for us now." Her honest tension: "There are so many new tools. It's nice to see, but it's overwhelming."

You can't ever blindly trust this tool.
— Natalie Denman, CIA, Audit Supervisor, Flowserve Corporation

Both things are true simultaneously. AI is genuinely making audit more efficient and the proliferation of tools creates its own burden.

DataSnipper is used by all Big Four for audit evidence extraction directly in Excel — if the Big Four have standardized on it, the mid-market will follow. MindBridge AI has improved anomaly detection rates by 60% by analyzing 100% of transactions rather than samples. PwC is targeting a fully AI-integrated audit platform by end of 2026.

The Stanford seniority finding lands hardest in audit: senior auditors who treat AI as a collaborator and step in when system confidence drops see larger performance gains than junior staff who accept outputs uncritically. Professional skepticism — the foundational auditing standard — is exactly the human layer AI cannot replicate, and what the PCAOB's September 2025 guidance development is insisting must remain. Your moat isn't the ability to foot a column. It's your judgment about whether the number makes sense. Develop AI verification as an explicit, practiced skill. The auditors who do will outperform those who either avoid AI or blindly trust it.

If You Work in Tax

Some of the highest automation claims in any accounting function — and also the clearest evidence that credential-based expertise is becoming more valuable, not less.

AI cuts standard return preparation time by 50-70% and reduces manual data entry errors by 75%. Thomson Reuters' 2025 survey found 21% of tax firms already using generative AI, with 53% planning to. This is happening at the return-preparation layer now — W-2 extraction, 1099 intake, preliminary schedules. Expect most firms to deploy AI for first-pass return prep by end of 2026.

Barrett Young, CPA, small-firm partner at GWCPA, uses Ask Blue J as "the tax manager I always wish I had, where I could ask any question and not get judged." His accuracy estimate is roughly 95% — meaning about 1 in 20 facts may need correction, enough to be extremely useful, not enough to use without verification. "I'm always telling my staff, 'Double-check that.'" Ask Blue J is the most practitioner-validated tool for individual tax professionals in this space. It's subscribable by individuals without waiting for employer adoption, which makes it the strongest recommendation for anyone who wants to move on AI research now. The honest caveat: numerical calculations remain a weak point across all AI research tools. Thomson Reuters CoCounsel Tax is the enterprise alternative worth knowing if your firm is evaluating platforms.

Accordance (multi-agent tax research, Stanford AI Lab founders) and Magnetic (YC-backed, AI tax preparer for CPA firms) represent the next wave. Neither is individual sign-up yet, but these are the names to watch over the next 12 months.

As the commodity prep layer erodes, the credential layer becomes the differentiator. Enrolled Agents and CPAs can represent clients before the IRS, sign returns, and provide legally privileged advice — things no AI tool currently does. The erosion of data-entry tax work strengthens the credential's value, because what remains is the judgment-intensive, relationship-intensive, credential-required work. If your tax work currently overlaps significantly with what AI is automating, credential investment has the highest near-term ROI of anything on this list.

The Skills That Actually Matter Right Now

AI skills mentions in accounting job postings rose 67% in the past year — 30% of listings now require them, up from 18% in January 2025. Thirty-one percent of CFO job listings now mention AI or machine learning. Employers are already repricing the talent market. Here's what to actually invest in, ranked by near-term return.

Verification discipline is first, and it's free to develop. The Stanford finding is specific: senior accountants who verify AI outputs outperform junior staff who accept them. Denman's revenue-figure incident is the clearest illustration of what happens when that discipline slips. The practical skill isn't asking "is this right?" — it's asking "how would I check this against source data?" That's a professional habit, not a technical skill, and it costs nothing but attention to build.

Iterative prompting is second, and it's learnable in days. Don Tomoff, CPA, 30-year veteran, eliminated $150/hour outsourced coding using ChatGPT since January 2023. His method: "You don't want to eat the elephant in one effort." Break tasks into stages, prompt for each stage, iterate. Brianne Smith, CPA/PFS, PhD, frames it as four phases: "Ask the question, master the data, perform the analytics, share the story." These are the same insight — AI performs best when given narrow, well-defined tasks. This is what "AI skills" means in practice when employers list it in job postings.

Data analytics and Python/SQL basics are the longer investment, but they're increasingly necessary if your role is moving toward FP&A or advisory. Thomson Reuters explicitly names these as the rising technical skills for accountants. DataCamp is the most practical individual-accessible path — its accounting and finance tracks are built for working professionals who need hands-on skills, not theory. Disclosure: this is an affiliate link. This is a 3-6 month investment; prioritize verification and prompting skills first, then come back to analytics. If you want the strategic foundation for thinking about human-AI collaboration before diving into tools, Ethan Mollick's book Co-Intelligence — written by a Wharton professor — is the clearest framework available for how to position yourself as the human layer that makes AI worth using.

The advisory pivot deserves a sentence: 43% of accountants now spend more time on advisory than compliance, and advisory rates run 40-60% higher. The technical skills only pay off if you can translate AI-generated analysis into client-facing recommendations. That communication ability is the demand-side skill that no tool teaches.

Where This Goes From Here

Three honest assessments based on your situation.

If you're in bookkeeping or AP and your work is mostly data entry, the pressure is real and moving faster than most articles admit. The path forward is becoming the person who operates and supervises the tools — not fighting them. Start with whichever close or AP tool your current employer uses. If they don't use one yet, that conversation is worth starting.

If you're in audit, your professional skepticism is the moat, but only if you actually apply it to AI outputs. The single most career-protective move is developing explicit verification habits now, before your firm deploys more AI and expects you to supervise it competently.

If you're in tax, the credential is your strongest differentiator as the commodity prep layer automates. Pair it with one AI research tool you can use personally today — Ask Blue J is the practitioner-validated starting point.

Two signals worth tracking over the next 12 months: whether the PCAOB formalizes its AI risk management guidance (when it does, firms will move fast, and accountants who already understand AI verification will have an advantage in those conversations), and whether Basis — now valued at $1.15B after its February 2026 Series B — expands its agentic AI beyond its current Big Four client base. That's the signal that autonomous accounting has crossed from early-adopter to mainstream. Neither requires action today. They require awareness. Only 37% of accounting firms currently invest in AI training, which means self-directed learners have a real and specific edge right now — one that won't last indefinitely.

Recommended Tools & Resources