In April, EY announced it was embedding agentic AI directly into its audits, a multi-agent system wired into EY Canvas, the platform that processes more than 1.4 trillion lines of journal-entry data a year, with the stated goal of supporting every end-to-end audit task by 2028. It was the clearest signal yet of where the largest firms are taking this technology.

It also lands in the same stretch during which those same firms have been quietly cutting the jobs that turn graduates into accountants. Two weeks ago we argued that AI’s real danger to the profession was not that it takes jobs, but that it eats the entry-level work that trains the next generation. The Big Four are now running that experiment at full scale, in public.

Start with what is being built, because this is not pilots anymore. EY’s EY.ai Agentic Platform, created with NVIDIA, launched in March 2025 with an initial 150 AI agents supporting 80,000 of its people across tax, risk, and finance. The April rollout pushed that framework into assurance itself, a multi-agent system embedded in EY Canvas and built on Microsoft’s Azure, Foundry, and Fabric stack, meant to handle end-to-end audit work by 2028. The March launch alone was pitched at surpassing 3 million tax compliance outcomes and reworking 30 million tax processes a year. These are production targets, not demos. Deloitte unveiled Zora AI, also built on NVIDIA, a suite of ready-to-deploy agents for finance and other business functions. KPMG committed $2 billion to AI with Microsoft back in 2023 and signed a global alliance with Anthropic this May to put Claude in front of its workforce. PwC has run GL.ai, an anomaly-detection tool for the general ledger, inside its audits for years.

Across the four, the AI spend runs into the billions. The point worth holding onto is what these firms are: they audit most of the world’s large companies and train a large share of its accountants. They are now wiring agents into the core of the work, not the edges of it.

Now the half that got far less attention. Over the past two years the Big Four have been shrinking their graduate intake. In the UK, where the numbers are clearest, KPMG cut its trainee class from 1,399 to 942, a 29 percent drop. Deloitte cut 18 percent, EY 11 percent, PwC 6 percent. Graduate job postings across the four fell 44 percent in a year. EY has delayed graduate start dates three years running. These are UK figures. The pull-back is not a British quirk, though: the same incentives, AI plus cost pressure, point the same way across the major markets, and the UK is simply where it has been counted most cleanly.

The reporting covering it is not coy about the cause. The compliance checks, document summaries, and routine memos that filled a new hire’s first years are now handled, in part, by the same kind of agents the firms are buying. The work that used to train people is the work being automated.

It is not purely an AI story, and the honest version says so. The same period saw the firms move headcount offshore, growing in Pakistan, Indonesia, and Malaysia while trimming in the UK, Germany, and the Netherlands. KPMG cut UK staff around 7 percent while adding roughly 10 percent in Pakistan; EY contracted about 6 percent in Germany and grew 7 percent in Indonesia. AI and offshoring are squeezing the entry tier from two directions at once. But the firms and the people covering them put AI at the center of why the junior headcount is falling.

Here is the part that matters beyond four firms. The Big Four are the profession’s training academy. A large share of the experienced accountants now working in industry, in government, and at every mid-market firm started in a Big Four graduate program, spent three to five years in the grind, and built their judgment there before moving on. That pipeline does not just staff the Big Four. It supplies the whole profession with seasoned people.

So when the firms that train the most accountants automate the training work and hire fewer trainees in the same year, they are not only cutting their own costs. They are throttling the source of experienced talent that every other employer draws from. Two weeks ago we called this the second shortage: not a lack of new entrants, but a lack of people with a decade of pattern recognition who can look at a set of books and catch a confident mistake. The Big Four are where most of those people were made. If the academy stops minting them, the shortage of judgment the profession already feels gets manufactured at its origin, on purpose, by the institutions best positioned to see it coming.

And the damage is the kind nobody sees for years. A firm can run perfectly well for a while on the seniors it already has, leaning on people trained a decade ago, while quietly hiring fewer juniors beneath them. The books still close, the audits still ship, the margin even improves. Then those seniors retire or leave, the thinner cohort behind them never got the reps to replace them, and the gap opens all at once, with a five-year lead time to fix it. Automating the partner tier would be obvious and alarming. Automating the trainee tier is cheap, nearly invisible, and exactly as consequential, only later.

The firms would tell a different story, and parts of it are fair. They are not abandoning graduates, they say; they are changing the job. The pitch is that a new hire freed from rote compliance work can do higher-value analysis sooner, supervise the agents instead of imitating them, and reach real judgment faster than the old apprenticeship allowed. There is also genuinely new work appearing. As clients push AI into their own books, someone has to assure that those systems are reliable, and the firms are standing up AI-assurance practices to do exactly that. That is potentially a real new rung on the ladder, validating other companies’ AI rather than crunching their ledgers, and it is precisely the judgment-heavy work a profession short on judgment should want its juniors cutting their teeth on. If the Big Four reinvest the hours AI saves into training juniors on harder problems, the pipeline survives in a better shape than before.

The question, the same one we raised two weeks ago, is whether they will. Cutting a graduate class is a decision a firm makes this quarter and sees in the margin immediately. Rebuilding how juniors learn is slow, expensive, and pays off in a decade. The incentives do not favor the patient choice, and the firms answer to partners who are paid annually.

There is a way to tell which story is the real one, and it is worth watching for. If the firms are genuinely reinventing the junior job, it will show up in what they say about training, not just hiring: new apprenticeship models, juniors put on judgment work in year one, real investment in developing the smaller cohorts they do take. If all you see is headcount falling and agent counts rising, with the training budget quietly following the graduate budget down, it is not reinvention. It is a cut with a better press release.

For the smaller and mid-market firms most of our readers run or work in, there is a concrete implication. The old strategy, let the Big Four train people and hire them away at year three or five, has worked for decades. It is getting less reliable. The academy is shrinking its output and automating the years that produced the most learning. If your plan for senior talent in 2030 is to poach Big-Four-trained managers, the supply you are counting on is being throttled right now, upstream of you, by decisions you do not control.

The defensive move is the one we laid out last time. Build your own training deliberately. Use AI to get your juniors to judgment-level work faster, have them review and interrogate agent output rather than do the rote work by hand, and keep a human accountable for everything that leaves the building. The firms that treat their bench as something to develop, not a cost to cut, will be the ones that still have senior people to sign the work in a few years. The Big Four can afford to run this experiment and absorb a bad result. A twelve-person firm cannot.

The Big Four have placed the largest bet in the profession: automate the junior tier, and trust that judgment will still somehow get made. If the bet pays off, they will have built leaner, faster firms. If it does not, the bill will not stay inside their walls. It will land on every employer that quietly assumed the academy would keep graduating the people they all depend on.

— Footnote

Footnote is an independent publication. It is not professional accounting, tax, or legal advice. Our analysis and opinions are based on the company announcements and reporting linked above. Graduate-intake figures are UK trade-press reporting; firm-level details are current as of June 2026 and subject to change. We have no consulting relationships with any vendor or firm named in this article.