The accounting profession has spent three years talking about two things and almost never at the same time. One is the shortage: there are not enough accountants, the pipeline is thin, and firms cannot hire fast enough. The other is AI: the tools can finally do real accounting work, and the vendors are selling them as the answer to the first problem.

Put the two stories in the same room and a problem appears that almost nobody is pricing in. The work AI is best at is the work that used to turn a new graduate into an experienced accountant. The cure eats the thing it is meant to cure.

Start with the numbers, because they are not subtle. More than 300,000 US accountants and auditors left the profession between 2019 and 2022, roughly a 17 percent drop from the peak, on Bureau of Labor Statistics figures. The pipeline behind them is thin. The number of accounting graduates has fallen about 20 percent since 2010, and 2022 brought a 7.4 percent single-year drop in degrees awarded, the steepest in three decades. The AICPA stopped calling it a shortage and started calling it a pipeline crisis, and the workforce that remains is aging toward retirement. On the hiring side the squeeze is concrete rather than abstract: roles stay open for months, and busy season increasingly runs short-staffed.

The profession is not sitting still. In May 2025 the AICPA and NASBA approved a new licensure pathway that lets a candidate qualify with a bachelor’s degree, two years of experience, and the exam, an alternative to the long-criticized 150-hour rule that effectively priced a fifth year of tuition into becoming a CPA. Dozens of states have moved since. PwC doubled its CPA exam bonus to $10,000 in late May, following EY. And there is a real green shoot: the AICPA reported that undergraduate accounting enrollment rose about 12 percent in fall 2024, the first meaningful uptick in years. The front door is being pried open. That part is genuine good news.

Now bring in AI, because the timing is what makes the pitch irresistible. Every product we have covered this spring is sold, explicitly, as a way to do more accounting with fewer accountants. Ramp’s Stack does not assist with the monthly close, it runs it, coding transactions and posting entries, and Ramp is selling it into the thousands of firms already on its rails. The Big Four are not dabbling either; they are pushing AI across audit, tax, and compliance and committing real money to it. The logic is clean: if you cannot hire the people, automate the work.

And the work AI does well right now is exactly the work nobody can staff and nobody enjoys: the reconciliations, the tick-and-tie, the transaction coding, the routine compliance checks. On paper, a labor shortage and a wave of automation aimed at the unstaffable grind look like a problem and its solution arriving at the same moment.

The routine work AI is best at is the same work that trained junior accountants into senior ones.

A new graduate did not learn judgment in a classroom. They learned it doing thousands of reconciliations, tying out thousands of numbers, and slowly noticing the patterns: the things that look wrong, the questions worth asking, the entry that does not smell right. The grind was the curriculum. The Journal of Accountancy put the question plainly in March: how will accountants learn new skills when AI does the work? The Big Four are already automating, as Fortune reported in May, the very entry-level tasks that served as the training ground, and the people raising the alarm are blunt about it: learning by doing stops working when there is nothing left to do.

Follow that forward and the shortage does not get solved. It moves. A firm that uses AI to avoid hiring and training juniors this year has fewer mid-level accountants in three years and fewer senior reviewers in eight. And the senior reviewer is not incidental. In our first two editions we argued that the human who checks the AI’s work and signs their name to it is the entire job now. That person is produced by the exact pipeline AI is hollowing out. You cannot automate the bottom of the career ladder and expect the top of it to keep filling on its own.

It helps to split the problem in two. There is a shortage of new entrants, the people coming in the front door, and the profession is genuinely attacking that one with cheaper licensure, recruiting bonuses, and an enrollment bump. Then there is a shortage of experienced accountants, the people with a decade of pattern recognition who can look at a set of books and know what is off without being told where. That second shortage is the one that actually hurts, because judgment cannot be hired in bulk or taught in a weekend.

AI does almost nothing for the second shortage, and it may quietly deepen it. You make an experienced accountant the slow way, by running a junior through years of the exact work AI now absorbs. So fix the front door all you want. If the hallway behind it stops developing anyone, the profession lands a few years out with more licensees and fewer people who can do the hard part. The headline number, total accountants, might even recover. The number that matters, accountants who can catch a confident mistake, would keep falling.

The optimistic read deserves a real hearing, because it might be right. The junior job does not have to vanish. It can change. Instead of grinding reconciliations, a new accountant could spend year one reviewing AI output, stress-testing its assumptions, and exercising judgment far earlier than the old apprenticeship allowed. That is arguably better training, and faster. A junior who spends year one asking why the model coded something a certain way, and where it is most likely wrong, may build judgment quicker than one who spent that year keying in receipts. The skill was never the data entry. It was the noticing.

The profession is visibly trying to build that version. The licensure reforms, the enrollment uptick, and programs like Intuit’s new pipeline initiative for training accountants in an AI-first world all point to a system trying to rebuild the on-ramp rather than abandon it. If firms deliberately redesign how juniors learn, the pipeline survives in a new shape, and a smaller number of better-trained people may be exactly what a shrunk profession needs.

The whole bet sits inside that one word, deliberately. It assumes firms will spend money training juniors to do judgment work when the cheaper short-term move is to not hire them at all. Left alone, most firms optimize for the quarter, and the quarter has never once rewarded developing someone you did not strictly need to hire yet.

Treat AI as a reason to change what juniors do, not a reason to stop hiring them.

Use it to get a new accountant to judgment-level work in months instead of years, by having them review and interrogate AI output from day one rather than do the rote work by hand. Build the training that the reconciliation grind used to provide on purpose, because it will not happen by accident anymore. In practice that can mean pairing every junior with both an AI and a senior, and having the junior explain the AI’s output to the senior rather than the other way around. The explaining is where the judgment forms. Keep a human signing the work, which is the same thing we have said since edition one. And run the math on the long game, not just the quarter: the firm that uses AI to quietly stop developing people will look more profitable for about two years, and then find it has automated its own succession plan. And measure the right thing. A firm watching only this quarter’s realization rate will cut juniors and look sharp doing it. A firm asking whether it can still staff a hard engagement in 2030 will make a different call.

The shortage is what made AI irresistible to this profession, and the tools are genuinely good at the work that is hardest to staff. But “fewer people, more AI” only holds if you keep making new accountants, and new accountants were made by the work AI now does.

The firms that figure out how to train judgment inside an AI-first workflow will own the next decade. The ones that simply cut the bench will spend it wondering where all the senior people went. The first shortage was an accident of demographics and a bad incentive. The second one, if it comes, will be a choice.

— Footnote

Footnote is an independent publication. It is not professional accounting, tax, or legal advice. Our analysis and opinions are based on the data, professional-body reporting, and sources linked above. Figures are current as of June 2026. We have no consulting relationships with any vendor named in this article.