Mercor, a company that builds tests to measure whether AI can actually do real jobs, just ran an experiment that matters to anyone who manages a back office. It hired twelve licensed accountants, each with several years of experience, and had them complete simplified bookkeeping tasks side by side with AI models. The AI won on every measure that counts: it finished faster, made fewer mistakes, and cost far less per task.
That alone is notable, but the harder version of the test tells the fuller story. Mercor also built a much bigger set of 160 realistic accounting tasks, based on ten made up companies, written by accounting professionals with years of real experience. On that harder test, the best model today solves a little under two thirds of the grading points, and well over half of the tasks are never fully solved by any model at all.
So the honest read is this: AI has gotten very good at the kind of bookkeeping that is really just following rules and hunting for the right number in a pile of documents. It is still weak at the kind of accounting that requires calling a client to ask a question, checking with a colleague down the hall, or remembering how a particular company has always handled a strange expense line. Mercor says that is exactly why accountants are not going away, even as the routine parts of the job get automated.
What makes this worth paying attention to is the timing. The accounting profession in the United States is dealing with one of its worst staffing shortages in decades, with fewer students choosing to study accounting and fewer people sitting for the CPA exam. At the same time, large accounting firms in the United Kingdom have already started cutting how many graduates they hire each year, pointing directly at AI handling the entry level work those graduates used to do.
Junior accounting jobs are not just cheap labor, they are training grounds. People used to spend their first few years doing the structured, repetitive tasks that AI is now better at, and that is how they built the judgment and client relationships that later made them senior accountants. If firms keep shrinking that entry rung because AI can do it cheaper, the supply of experienced accountants who can supervise the AI gets smaller too, right when the need for that supervision is going up.
For any business that relies on outside accountants or runs its own finance team, the near term move is not to cut headcount, it is to change what junior staff spend their time on. Let AI handle the reconciliations and the data entry, and point the humans you already have at the judgment calls, the client conversations, and double checking what the machine produced. The firms that work out that split first will run cheaper books without losing the people who will eventually need to run the whole department.