Will AI Replace Accountants? Not the Way Anyone Expects

A startup called Mavi came out of stealth on Monday with $4 million in seed funding and a bet that sounds backwards: the AI boom is going to create demand for accountants, not erase it. TechCrunch reported that the company, founded in 2023 and run by co-CEOs Molly Liu and Aman Puri, has built a marketplace connecting US companies with AI-proficient finance and accounting talent. Harlem Capital led the round, which closed last year.

So, will AI replace accountants? Liu’s answer is more specific and more uncomfortable than a yes or a no. “AI will automate away many entry-level roles in finance and accounting, which will further exacerbate the talent shortage at the mid-level experience level, given there will be less entry-level talent growing into the mid-level over time.”

Read that twice. The claim isn’t that accounting goes away. It’s that the bottom rung of the ladder gets sawed off, and a few years later nobody can find anyone standing on the third rung, because the third rung was only ever reachable from the first.

That’s a very different problem than the robots-took-my-job version. It also changes where the jobs are and how you find them.

What AI is actually taking

The pattern Liu describes is already visible in payroll data. Stanford’s Digital Economy Lab analyzed millions of anonymized records from ADP, the largest payroll processor in the country, covering late 2022 through mid-2025. The researchers found a 13% relative employment decline for workers aged 22 to 25 in the most AI-exposed occupations. Older workers in those same occupations held steady or grew.

The detail that matters most got skipped in most of the coverage. The decline wasn’t uniform across AI-exposed work. It showed up where AI automates a task, and not where AI augments one. In the study’s words, “entry-level employment has declined in applications of AI that automate work, with muted changes for augmentation.”

Now list the work a first-year accountant does. Bookkeeping. Reconciliations. Coding accounts payable. Pulling numbers into a template so someone senior can look at them. That list is close to a working definition of automatable.

The researchers’ explanation for why young workers absorb the hit is that AI is good at replacing codified knowledge, the textbook material a graduate shows up with, and bad at replacing tacit knowledge, the judgment that accumulates only by doing the work badly for a while and getting corrected.

That’s the trap. The tacit knowledge used to come from grinding through the codified work. Remove the grind and the on-ramp goes with it.

Worth noting what the same study found on the other side of the line. Where AI augmented a task rather than automating it, such as supporting problem-solving or checking accuracy, young-worker employment held steady and sometimes rose. The technology isn’t the variable. How a given company chooses to deploy it is. Two finance teams running the same software can end up with opposite hiring patterns, which is one more reason company-level research beats scanning job titles.

Will AI replace accountants, or just the route to becoming one?

The supply side was thin before any of this started. The AICPA’s 2025 Trends Report found that students earning a bachelor’s or master’s degree in accounting fell to 55,152 in the 2023-24 academic year, down 6.6% from the year before. Master’s degrees in accounting and taxation dropped about 15%, to 14,335.

The CPA exam numbers tell a similar story with an asterisk. 42,626 new candidates entered the pipeline in 2023, the highest count since 2016. In 2024 that fell to 28,082. Part of that swing is an artifact, since 2023 was the year before a major exam overhaul and those years always see a rush to beat the change. But 28,082 is still a small number set against a profession that needs replacements.

There are signs of a rebound. The same AICPA report cites National Student Clearinghouse data showing two consecutive semesters of 12% year-over-year growth in accounting enrollment during 2024-25, and 16,448 new CPA candidates entered through the first half of 2025. Jan Taylor, the AICPA’s academic-in-residence, called the slowing rate of decline encouraging.

Stack the two trends against each other anyway. Fewer people entering the profession, and fewer entry-level seats for the ones who do. Liu’s framing of the underlying problem is blunt: the US accounting profession is shrinking, and the people still in it aren’t necessarily fluent in AI.

Which means the useful question isn’t whether the profession disappears. It’s who will be left in five years holding both the fundamentals and the ability to run the tools. The honest projection is a squeeze in the middle. Plenty of senior people. A thinner-than-usual layer with four to eight years of experience, because the pipeline that produced them dried up at exactly the moment demand for them went up.

The role is being redefined, not deleted

Liu’s definition of an AI-proficient accountant is worth spelling out, because it describes a job that doesn’t exist in most HR systems yet. Strong finance and accounting fundamentals come first. On top of that: the ability to direct AI tools to do work, the judgment to catch what the AI gets wrong, the skill to integrate LLMs into finance and accounting workflows, and experience with AI-native ERPs.

Compare that to what she calls a traditional accountant, who “has the accounting fundamentals but only has a basic understanding of how to incorporate AI into finance and workflows or how to review [and] manage AI tools within finance.”

The gap between those two people isn’t a credential. Nobody issues a certificate for it. It’s a working habit.

Meanwhile the pressure moves upward. Liu’s observation is that the more AI tools a company adopts, the harder senior roles get, not easier. “They are increasingly expected to add more of their reasoning skills,” she said. And: “Finance teams are now being pushed to do more with less.”

This is the inverse of the automation story people usually tell. Volume work shrinks. Judgment work expands, because somebody now has to review output from a system that generates plausible wrong numbers very fast. Anyone who has hunted a $3,000 variance through a reconciliation a machine produced in four seconds knows the review is not the easy half.

For anyone already in finance or accounting, that’s the real career instruction. Not “learn AI” in the abstract. Get specific: which tool, in which workflow, catching which category of error.

The postings haven’t caught up to the work

When a role changes faster than the vocabulary for it, job postings are the last place the change shows up.

Postings get written by copying last year’s posting. They get slotted into a requisition system under a title that has to match an existing salary band. They get approved by people who weren’t in the room when the work changed. By the time “Senior Accountant, AI-Enabled Close” exists as a standardized title with a pay range behind it, hiring for that work will have been happening for three years through other channels.

Look at what Mavi is actually selling. Not a job board. A marketplace with more than 3,000 accounting professionals on it, aimed at full-time mid-level and senior placements, handling the cross-border contracts, legal work, compliance, and payroll so US companies can hire AI-proficient finance talent anywhere. Enterprise clients include the personal care company Athena Club. Liu came out of Ramp, Lyft, and Dropbox, which means she watched finance teams get rebuilt around software from the inside.

Companies are paying a venture-backed intermediary to find these people. Take that as a signal. It means the default method isn’t working, and the default method is posting a job and reading what comes in.

The same thing happened with growth marketing around 2013 and data engineering around 2017. Both were hired through referrals and operator networks for years before the title stabilized enough to post. The hidden job market isn’t a conspiracy. It’s the lag between when work changes and when HR gets a word for it.

Go to the person who owns the work

If the job you want doesn’t have a stable title yet, every hour spent refreshing job boards goes toward searching for a string nobody has written.

Direct outreach is the unglamorous alternative, and it works better here than anywhere else. Identify the companies where the work already exists, whether or not a req is open. In finance that’s the controller, the VP of Finance, the accounting manager, or the CFO at a company small enough that the CFO still reviews the close. Those people know they have a gap long before a posting exists, because they’re the ones covering it at 9pm.

Then say something specific. Not “I’m passionate about the intersection of AI and finance.” Closer to: you’ve run a month-end close inside an AI-native ERP, here’s the category of error you caught that the system didn’t, here’s what it did to your close timeline. A message like that isn’t competing with 400 applications, because it never enters the application pile.

If you’re early in your career and the first rung really is gone, the same logic applies with more urgency. The experience you can’t get from a seat you can’t land has to come from somewhere else: a small company where one person does the whole close, contract or fractional work, a nonprofit that needs books cleaned up. None of that is posted either. All of it is reachable by asking a specific person a specific question.

It’s uncomfortable. Cold outreach feels worse than applying, which is precisely why the pile is 400 deep and the inbox isn’t. That discomfort is the whole arbitrage.

The research is the hard part: working out who actually owns the work, finding something real to say about their company, and writing a message that doesn’t read like a template. Angld.AI does that pipeline for you. Paste a job posting, and it identifies the decision maker, researches them, and drafts a personalized outreach message in about 60 seconds.

So, will AI replace accountants? No. But it is quietly rewriting who gets to become one, and the people who notice first won’t be finding those roles on a job board.