How AI Will Replace Tax Preparers: The Intuit Cut and the End of Entry-Level Accounting

There is a particular kind of announcement that an industry makes when it has decided, quietly, that the entry-level version of itself is no longer worth hiring. Intuit made it on May 20, 2026. The press release said the company was cutting seventeen percent of its workforce, about three thousand jobs, to "accelerate AI-powered product development." The same release reported ten percent revenue growth in the quarter just closed. Intuit was not shedding workers because it was failing. It was shedding workers because the work the bottom of its org chart had been doing — tax-prep support, simple return review, customer assistance through tax season, junior accounting backoffice — was now being done by a model.

The Intuit cut is the cleanest single data point in a pattern that has been building since late 2024 and accelerated through the 2026 tax season. The Big Four reduced graduate openings forty-four percent year over year. EY delayed graduate start dates a third year in a row, with the 2025 hiring cohort not actually beginning work until March 2026. KPMG cut graduate intake by nearly thirty percent in some regions. TurboTax now lives inside Claude and ChatGPT through the Intuit-Anthropic partnership announced in the same May 2026 window — meaning the most common entry point into a tax return for an American household is now a conversation with a model, not a phone call to a CPA's office.

This article is the HAR analysis of what the bottom three rungs of the tax-and-accounting ladder look like as they get sawed off, what the top three rungs will look like as the work stops flowing up from below, and what the workers in the affected jobs need to know now that the canary has finished singing.

The Occupation Profile

The US Bureau of Labor Statistics counts approximately 180,000 tax preparers in the country — narrowly defined as the workforce whose primary function is preparing and filing personal and small-business tax returns. Median pay sits around $48,000 a year. The work is intensely seasonal: roughly seventy percent of the total annual hours worked happen between mid-January and mid-April. Roughly forty percent of the workforce is employed by the three major retail tax-prep chains (H&R Block, Jackson Hewitt, Liberty Tax), with the remainder distributed across independent practices, small accounting firms, and seasonal storefronts.

Adjacent to this narrow category sits a much larger one: the entry-level accounting workforce. The BLS counts approximately 1.4 million accountants and auditors in the US, of whom roughly two hundred thousand are in their first three years of practice — what the Big Four would call "associates" and the IRS would call "preparers under direct supervision." This is the bottom of the accounting career ladder, and it is the population that does the work most directly comparable to the work an AI can now perform end to end.

These numbers are the denominator. The pieces below are about what fraction of each row gets displaced over the next twelve to thirty-six months, and why the displacement curve is not linear.

What AI Is Now Doing End-to-End

Three years ago, "AI in tax prep" meant document OCR plus a rules engine plus a chatbot wrapper. The output was a draft return that a human had to review line by line because the model would confidently classify a 1099-NEC as wage income or hallucinate a state-specific credit that did not apply to the taxpayer's situation. The human review was load-bearing — it caught real errors at a rate high enough that the firm could not have shipped without it.

What changed in 2025-2026 is that the model-level capability stopped being the bottleneck. The current generation of frontier models, when given a shoebox of source documents and a structured eval framework on top, can produce a return that beats the human-only baseline on accuracy in controlled studies. The Intuit-Anthropic case study published in the same May 2026 window reports that the TurboTax-Claude pipeline now processes forty-four million returns annually and that the human review rate has fallen from roughly one in three returns in 2023 to roughly one in twelve in 2026. Eight out of nine returns now ship without a human touching the output.

Eight percent is not zero, but it is also not the same eight percent it was a year ago. The returns that still get human review in 2026 are not the simple W-2 plus mortgage interest plus child tax credit returns that dominated the review queue in 2023. They are the multi-state, multi-entity, multi-jurisdiction returns where the model still flags low confidence on specific line items. The bottom rung is gone. What remains is the exception-handling layer.

The corresponding labor effect is the one the rest of this piece is about: when the model handles eleven-out-of-twelve returns without a human, the firm needs roughly one-twelfth of the human-prep workforce it used to. The math is brutal because it is exactly the math.

The Intuit Cut, Read Literally

The Intuit announcement on May 20, 2026 cited "accelerated AI-powered product development" as the rationale for the seventeen percent workforce reduction. Read literally, that means: the company is reducing headcount in roles whose marginal contribution is below what the AI-augmented version of the same role produces. Specifically — and Intuit was careful not to say this publicly, but the role categories in the WARN Act filings make it clear — the cuts concentrated in tax-prep support staff, content moderation reviewers for AI-generated tax explanations, junior product managers in the consumer tax line, and the seasonal expansion staff that Intuit historically scales up between January and April.

The categories above add to roughly three thousand. The two largest — tax- prep support and seasonal expansion — are the rows that map most directly to the "tax preparer" occupation in the BLS data. Intuit is not eliminating the senior tax-CPA function. It is eliminating the layer of work that the senior function used to delegate to.

This is the structural shape of the displacement. The senior practitioner is more valuable than ever per hour; the work that flowed to the senior practitioner from junior preparers has been intercepted by the model. The senior practitioner now reviews the model's output, escalates the exceptions, and bills the same hourly rate against a smaller team. The result is the highly leveraged senior tax practice that the industry has been gesturing toward for two decades but never operationalized at scale.

The Big Four Entry-Level Collapse

The Big Four story is the same story at a different scale. The reported forty-four percent year-over-year drop in graduate openings is the aggregate; the firm-by-firm breakdown is more revealing.

EY's fifty-one percent reduction is the largest, and EY is also the firm that has most publicly pushed back graduate start dates. The 2025 graduate cohort at EY was hired in autumn 2024 and originally scheduled to start in September 2025; the actual start date for that cohort, per internal communications that surfaced in trade press in March 2026, is now scheduled for the autumn of 2026. The firm is paying retainers — significantly below full salary — to keep the cohort attached while it figures out what work they will actually do when they show up.

That is not a temporary measure. That is an industry telling itself a story about how it will eventually use these workers, while the actual demand signal for their labor remains absent. EY's situation is the most public version of what all four firms are doing.

The work that the entry-level accountant historically did — bank reconciliations, journal entries, audit confirmations, footing tie-outs, basic 10-K assembly — is the work that AI now does without human involvement. Internal pilots at all four firms report tie-out cycle times falling from hours to minutes and audit confirmation work moving from a junior-staffed task to an unattended pipeline that the senior auditor spot-checks.

The Skill-Ladder Problem

Here is where the analysis gets uncomfortable. The traditional career path in tax and accounting was a ladder. You started as a preparer or junior auditor. You spent two to four years doing repetitive technical work, during which you absorbed the implicit knowledge — what a normal transaction looks like for a manufacturing client versus a SaaS client, how to spot when a balance is off in a way that suggests fraud versus a way that suggests sloppy bookkeeping, what kinds of questions a partner will ask when reviewing your work. After four years, you became a senior associate, then a manager, then a partner-track candidate.

The ladder worked because the repetitive technical work was the training data. The work the junior did taught the junior what work the senior needed to do.

When the AI does the repetitive technical work, that training-data layer disappears. The senior position still exists. The senior practitioner still needs to know what a normal manufacturing-client transaction looks like. But the path by which a fresh graduate became someone who knew that is gone.

The chart above is the skill-ladder problem in one frame. The associate row collapses fast. The senior row holds for a while but starts to bend down by 2026 because the firms recognize that without an associate pipeline they will not have enough seniors in three years. The manager and partner-track rows are most insulated, for now, because they are doing client-facing and judgment work that the AI cannot do unassisted. But the partner-track row in 2030 is filled by people who were associates in 2026, and there are no associates in 2026.

This is the meta-problem that the firms are trying to solve without admitting they have it. None of them has announced a serious plan for how to develop senior practitioners in a world without an entry-level training pool.

The Geographic Distribution of the Cuts

A useful test of whether a workforce reduction is genuinely AI-driven versus a generic cost-cutting exercise is whether the cuts cluster where labor was cheapest, or whether they cluster where the work was most automatable. Generic cost-cutting concentrates layoffs in high-cost geographies. AI displacement concentrates them where the work was most amenable to automation, which is largely independent of cost.

The Intuit reduction profile passes the AI-driven test. Cuts were distributed roughly in proportion to where the AI-replaceable work sat, not in proportion to where the labor was most expensive. Mountain View and San Diego — the company's two highest-cost geographies — were proportionally less affected than Bangalore and Cebu, where the seasonal tax-prep support work was concentrated. The pattern matches what the Big Four are doing: graduate intake reductions are largest in the regions where the entry-level tax and audit work was outsourced to in the first place.

That distribution is the smoking gun. The work that moved offshore in the 2010s was the work most legible to AI replacement in the 2020s. The offshore tax-prep operation was a stable target because the work was structured, repeatable, and rules-bound — the same properties that made it the first thing a model could absorb end-to-end. The offshore operations are being reduced faster than the domestic ones because they were where the AI-replaceable work was actually living.

The implication for the workforce question is that the displacement is not primarily a US labor-market story. It is a global story, with the specific countries that absorbed accounting-and-tax offshore work in the past decade now absorbing the largest workforce reductions. India alone has roughly four hundred thousand accounting-and-tax-prep workers in offshore captive centers servicing US firms; a thirty-to-forty percent reduction in that workforce is the silent counterpart to the Big Four graduate hiring collapse in the US and UK.

How the Displacement Mechanics Actually Work

The mechanics deserve a closer look because they determine which roles survive and which do not.

The AI capability that matters most is not "doing taxes." It is structured-information extraction from unstructured source documents followed by rules-engine execution against a knowledge base with the ability to flag low-confidence outputs for human review. That sentence describes about eighty percent of the work the tax-prep junior workforce was doing. It also describes about sixty percent of the work the entry-level audit and assurance workforce was doing. It describes much less of the work the senior advisory workforce was doing.

The reason the model can now do this work end-to-end and could not three years ago is a stack of three changes that landed together during 2024-2025.

First, document understanding got reliable. The 2024-2025 generation of multimodal models reads a stack of source documents — a shoebox of W-2s, 1099s, brokerage statements, K-1s, mortgage forms, charitable receipts — with accuracy on the line-item extraction task that is now better than a trained human preparer working at typical production speed. The error rate on a W-2 box-by-box extraction is under one in a thousand. The error rate on a moderately complex K-1 extraction is under three in a hundred. Both are below the comparable human-junior error rate.

Second, tool-use and structured output got reliable. The model that reads the documents can now write structured calls to a tax calculation engine, get answers back, and assemble the answers into a return. The "agentic" capability that was unreliable in 2024 is production-ready in 2026. The Intuit case study reports that the TurboTax-Claude pipeline runs an agentic loop with an average of seventeen tool calls per return, and that the loop reliability — the fraction of returns where the agent completes without a critical error — exceeds ninety-eight percent.

Third, eval frameworks got serious. The deployment risk for AI tax prep was always that the model would confidently produce a wrong return. The 2025-2026 generation of eval frameworks for tax-prep AI runs the candidate model against a corpus of approximately two hundred thousand reference returns with known-correct outputs and catches the failure modes before they reach customers. The firms that deployed AI tax-prep at scale without serious eval frameworks (there were several in 2023-2024) experienced expensive errors. The firms that deployed with serious eval frameworks (Intuit, H&R Block, several mid-tier chains) have shipped with error rates below the human-only baseline.

These three changes compound. The capability gap between an AI-enabled tax-prep pipeline and a human-only pipeline grew faster than linear during 2024-2026, and the gap is what created the workforce reduction.

The crossover point is somewhere in mid-2024 on the composite score. The acceleration after the crossover is the gap that drove the 2025-2026 workforce reductions. The flat human line is a feature, not a bug — the human-only baseline does not improve over time at the rate the model does because the bottleneck for the human is hours-per-return, not knowledge.

What Is Actually Left for Humans in Tax and Accounting

The work that remains in the post-2026 occupation has a specific shape. Five categories survive the displacement curve:

First, complex multi-jurisdiction tax planning. The tax code is deliberately knotted between federal, state, local, and international treaty regimes, and the planning work — sequencing transactions so a particular ownership structure produces a particular tax outcome over a five-year window — remains a senior-judgment activity. The AI can model scenarios; the partner picks which scenario to recommend.

Second, audit defense and IRS representation. When a return becomes a notice, becomes a correspondence, becomes a field audit, the work shifts from preparation to negotiation. The IRS field agent is a human; the defense is a human-to-human conversation with a particular adversarial dynamic. AI assists in pulling documentation, drafting positions, and scenario-planning the agent's likely next move, but the negotiating posture itself is human.

Third, forensic and fraud-detection work. The AI is excellent at flagging statistical anomalies in transaction streams. The follow-up investigation — interviews, document subpoenas, witness preparation — is where the human practitioner adds the most value per hour. The forensic practice is the part of the industry that is hiring through the displacement curve.

Fourth, niche-industry advisory. Cannabis tax compliance, oil-and-gas depletion accounting, multi-entity real-estate K-1 work, hedge-fund tax — these are niches where the volume is too low for general-purpose AI training to absorb the edge cases, and the cost of an error is too high for the AI-only pipeline to be deployed without senior oversight.

Fifth, the senior advisory and relationship layer. The CFO of a fast- growing private company is not going to take a strategic acquisition question to a chatbot. The CFO takes it to the partner who has been the firm's audit relationship for a decade. That conversation does not get automated. It gets cheaper because the partner now has a smaller team behind them and higher leverage per partner.

The composition above sums to one hundred percent of what is left. The absolute number is much smaller than the pre-2024 workforce. The distribution within that smaller number is more weighted toward judgment- heavy senior work than the historical mix.

The Honest Timeline

Three things will happen at staggered intervals between now and 2029.

In the next twelve months — between now and mid-2027 — retail tax-prep chains will continue to shed seasonal storefronts. H&R Block has already announced a fifteen-percent reduction in physical locations for the 2027 tax season. Liberty Tax is pursuing a similar consolidation. Jackson Hewitt is being shopped to private equity at an enterprise value that implies roughly half its 2023 valuation. The independent tax preparer will start to feel the squeeze in 2027 when the AI-enabled DIY tools expand into the kinds of small-business returns that have been the independent preparer's bread and butter.

In the eighteen-to-thirty-month window — mid-2027 through end-2028 — the Big Four entry-level collapse will start to flow into mid-level shortages. Firms that cut associate hiring in 2024-2026 will not have managers ready in 2028-2029. Some firms will respond by raising salaries sharply to compete for the limited remaining mid-level talent. Others will offshore the mid-level work that the AI cannot yet fully replace. A few will pioneer the "career re-entry" path that lets experienced practitioners from adjacent industries (finance, law, ops) re-train into the accounting senior role without going through the ladder.

In the thirty-six-month-plus window — 2029 onward — the durable post-AI profession settles into roughly what the chart above shows. Total workforce is approximately forty percent of the 2023 baseline. Median pay is meaningfully higher because the residual work is more senior. Geographic concentration shifts toward the major financial centers where the surviving advisory practices cluster.

The asymptote at forty percent is the bear case for the workforce and roughly the base case for the firms. Some firms will end up with a residual workforce above forty percent because their client mix is more complex; others will end up below because their client mix is more commoditized. The aggregate number sits where the chart shows.

What Workers in the Affected Roles Should Do

The honest advice depends on where in the career arc the worker currently sits.

For preparers and junior accountants in their first one-to-three years: the ladder you joined no longer goes where you thought it did. The realistic moves are (a) accelerate into the surviving niches — pick one of the five residual categories above and specialize hard, before the firms close the remaining slots; (b) pivot into the AI-adjacent functions that the firms are actually hiring for, including prompt engineering for tax workflows, eval design for accounting models, and the new "AI audit" role that several firms are piloting; or (c) leave the profession while the credential is still fresh and the adjacent industries (FP&A, internal audit, controls, treasury) still value an entry-level public-accounting background.

For mid-career associates and managers (years three through ten): the position you hold is more valuable than the model right now and the five-year outlook depends on whether you sit on the right side of the judgment-vs-execution line. If your day is mostly executing on tasks the AI can also execute, the next two years are when you transition to something more advisory. If your day is mostly client-facing planning and judgment work, you are in the part of the profession that is being leveraged-up by AI rather than replaced, and the right move is to deepen the relationship layer and accept a smaller team.

For senior partners: the strategic question is what the firm looks like in 2029 with one-third the headcount and the same revenue. The firms that figure that out early will keep their partners. The firms that defer the question for two more years will have a partnership crisis when the senior tier cannot replace itself from below.

What Firms Should Do

The firm-level prescription is uncomfortable because it requires saying out loud what the cuts have already implied.

First, stop pretending there is still a ladder. The marketing materials at all four Big Four firms still describe the associate-to-partner path as if it exists. It does not. The firms that will recruit best in 2027 are the ones that publicly redesign the entry point — not as "the first rung of the ladder you remember" but as "a two-year applied AI-audit fellowship that may or may not transition into the firm."

Second, build the mid-level on-ramp. The firms need a way to bring experienced practitioners from adjacent fields into the manager tier without the four-year associate gauntlet. This is partly a training problem, partly a credentialing problem (the AICPA will need to evolve the path-to-CPA in ways it is currently not moving), and partly a compensation problem.

Third, price the senior practice for the new model. Hourly billing against a smaller leverage pool is a worse business than fixed-fee advisory against a larger relationship pool. The firms that move first on the pricing model will capture the high-margin work the AI made possible.

The Credentialing Question Nobody Is Discussing

There is a quieter problem behind the workforce contraction that the trade press has not yet engaged with. The accounting profession's credentialing system — the CPA exam, the AICPA's ethics requirements, the state-level practice rules — is calibrated to a workforce that enters via the entry-level ladder and accumulates supervised hours en route to the credential. The standard path to the CPA in most states requires one hundred fifty academic credit hours plus typically one to two years of supervised work under a licensed CPA.

When the supervised-work portion of the entry-level role gets absorbed by AI, the credentialing path breaks in a quiet way. The candidate can still pass the exam. The candidate can still accumulate the credit hours. But the "supervised work experience" requirement — historically the part that taught the candidate how the profession actually operated — is now supervised work over an AI pipeline, which is not the same training experience.

The AICPA is moving slowly on this. Several state boards of accountancy have raised the question internally; none has published a revised credentialing framework. The window during which the existing credentialing path produces well-trained practitioners is the window in which today's third-year associates were hired, which closes around 2027-2028. After that, the firms will be producing senior practitioners from a candidate pool that has not had the supervised-execution-work training experience that the previous generation had. The profession will have to figure out what that does to the quality of senior practice in 2030 and beyond.

This is the kind of problem that does not show up in the workforce reduction numbers because it is a quality-of-the-residual-workforce problem rather than a quantity problem. The forty-percent-of-baseline workforce in 2029 is a different forty percent than it would have been in a world where the ladder still functioned, because the training data the senior practitioner accumulates is qualitatively different.

What This Means for Tax Software Companies

The Intuit cut is also a signal about the business model under the covers of consumer tax software. Intuit's revenue grew ten percent in the quarter the announcement covered. The workforce reduction was not a response to declining demand — it was a response to a shift in what the demand could be served with.

The dominant 2024 consumer tax-software product was a guided wizard plus a human-assistance escape hatch ("TurboTax Live"). The escape hatch was the human-prep workforce that the May 2026 cut largely eliminated. The 2026 product is a model that can converse with the user about their return, answer arbitrary questions about their specific situation, and complete the return without ever passing the session to a human. The product is simultaneously more capable and cheaper to deliver.

The strategic question for Intuit-and-peers is whether the post-2026 product can be priced at the historical levels or whether the same disintermediation that hit the human prep workforce now hits the software ARPU. The current evidence is mixed. TurboTax's average revenue per user is roughly flat year over year; the human-assistance upsell that used to convert a free-tier user to a paid-tier user is now a model-assistance upsell that converts at roughly the same rate. H&R Block has reported similar dynamics. If that pricing power holds, the post-cut Intuit is a higher-margin business than the pre-cut version. If it does not — if the next eighteen months reveal that users are not willing to pay $89 for a chat with a model the way they were willing to pay $89 for a chat with a human CPA — then the software business itself starts to compress and the cuts continue.

The margin lift in the chart is the case for the cuts from the company perspective. The risk is that the lift is one-time and the pricing erodes from 2027 onward as competition catches up.

The Broader HAR Pattern

The tax-preparer story is the cleanest current example of a pattern that this series has tracked across multiple occupations. The pattern has three stages:

Stage one: AI augments the senior practitioner. Productivity goes up. Headcount stays flat. The story is "AI doesn't replace anyone."

Stage two: AI replaces the junior practitioner. Headcount at the bottom falls. The story shifts to "AI is changing the entry-level role."

Stage three: AI replaces enough of the junior practitioner work that the senior pipeline starves. Headcount at the bottom does not recover. The senior tier holds steady or grows for a window, then begins to contract as the pipeline-feeding effect catches up. The story becomes "AI hollowed out the profession."

Tax preparers are in stage three as of mid-2026. Insurance underwriting is in stage two with stage three visible (see the insurance underwriting barbell analysis from May 21). Medical coding is in stage three. Legal document review is in stage two-and-a-half. The displacement curve is not symmetric across occupations and the timing varies, but the shape is the same.

The intervention point — the place where workers and firms have any leverage at all — is in stage two. By stage three the structure has already changed. The firms that started the work in 2024 are the ones that have a plausible answer in 2026. The workers who started specializing in 2024 are the ones with surviving careers in 2026.

The Intuit cut is a stage-three announcement. The Big Four hiring collapse is a stage-three pattern. The tax preparer occupation is no longer in a window where individual choices change the aggregate outcome. The aggregate outcome is set. What remains is which workers end up in the forty percent that survives and which firms end up running the practice that the survivors work for.

Further Reading

• The Underwriting Barbell — Insurance HAR analysis — the same hollowing-out pattern one industry over, with the barbell shape that tax-prep will end up matching.
• The Permanent Map — How American Work Hollowed Out 2015-2025 — the longer-horizon backdrop for what HAR Thursdays are tracking.
• My prediction on accounting-employment trajectory through 2027 — the falsifiable claim about where the workforce number lands.