How AI Is Changing CPA Training: Why Accounting Firms Need to Develop Judgment, Not Just Technical Skills

By Vincent Howard, CPA | Managing Partner, Howard, Howard and Hodges | SkillAbility for Accounting Firms

Last updated: 2026  |  13-minute read

Who I Am and Why You Should Listen

I’ve been in public accounting since 1990. I founded my own firm in 1993, merged it in 2001 to form Howard, Howard and Hodges, and grew it from three people to 50 staff across four locations and multiple states. Our firm was named PASBA Firm of the Year.

I’m a product of the old training model this article is about. I learned accounting by grinding — reconciliations, returns, the slow accumulation of judgment through thousands of repetitions and a senior’s red ink on my work. That model worked, in its inefficient way, because the repetitive work was always there to learn on. It isn’t anymore. Since 2020 I’ve built a development platform that more than a thousand accounting professionals across dozens of PASBA member firms have moved through — and watching how the next generation actually develops (and where they stall) convinced me the profession needs a fundamentally new training model, not a tweak to the old one. This article is what AI is forcing all of us to confront, and what to do about it.

The Old CPA Training Model Was Accidental

For a century, accounting trained its next generation almost by accident. A new graduate arrived, was handed the repetitive work nobody senior wanted — vouching transactions against invoices, keying data, reconciling accounts, preparing basic returns — and through the sheer volume of that work, slowly developed something no classroom teaches: judgment, pattern recognition, the instinct for when a number is wrong.

The AICPA’s own leaders describe the model plainly. Carl Mayes, who oversees CPA candidate quality at the AICPA, recalls that one of the first things he did as a young auditor was “vouch” — examining transactions and comparing them to supporting evidence to ensure they were genuine and accurate. That repetitive, time-consuming work was tedious. It was also the apprenticeship: doing it thousands of times is how judgment formed.

The repetitive work was never really about the work. It was the tuition. And that’s exactly the problem.

The grunt work that everyone wanted to automate was secretly doing two jobs: producing output, and producing accountants. Automate it away, and you solve the first job while quietly breaking the second. (We diagnosed this fully in the AI apprenticeship crisis; this article is about what to build in its place.)

AI Is Removing the Work That Used to Train People

That same vouching work Mayes learned on? Commercially available AI platforms now have vouching tools that do it automatically, saving enormous time. Across the profession, AI now collects documents, extracts data, validates completeness, flags exceptions, and produces review-ready work products — exactly the entry-level workload that historically trained new staff.

This is genuinely good for productivity. It’s also a development problem, because the work being automated was the work people learned on. As Accounting Today framed it, automation reduces repetitive tasks while simultaneously increasing expectations around analytics and client-facing judgment — meaning those entering the profession must deliver broader skill sets sooner. Higher expectations, less of the work that used to build the foundation for them. That’s the squeeze.

The profession’s response is unambiguous. The Journal of Accountancy reports that as entry-level tasks are automated, the focus of training must shift to judgment, simulation, and continuous upskilling — with passive lectures replaced by scenario-based, simulation-driven learning. The accidental apprenticeship has to become a deliberate one.

Why Technical Knowledge Is No Longer Enough

Here’s the shift that catches firms off guard: in an AI-enabled firm, the value of an accountant moves almost entirely from executing to judging. As AICPA’s Tom Hood puts it, “AI provides speed, scale, and analytical power. Accountants provide judgment, context, and integrity”. The machine does the task; the professional decides whether the machine got it right and what it means.

That reframes the entire skill requirement. CPAs are now expected to analyze the output generated by AI systems rather than perform the tasks manually — which requires strong critical-thinking skills to ask the right questions. The skills that matter now sit on top of technical knowledge, not in place of it:

• Critical thinking and professional skepticism — the ability to challenge an output rather than accept it
• Accounting judgment — knowing what’s reasonable, what’s anomalous, what needs a second look
• Client interpretation and communication — translating numbers into guidance a business owner acts on
• Escalation skills — recognizing the edge of one’s authority and competence
• AI supervision — evaluating, challenging, and governing automated work

Technical knowledge is the entry ticket. Judgment is the job. And judgment is exactly what the old accidental apprenticeship built and the new automated workflow doesn’t.

The New Risk: Faster Output, Weaker Understanding

This is the danger that should keep firm owners up at night, and it’s specific to the AI era: AI lets a junior produce work faster than they can understand it.

In the old model, slow work was a feature in disguise — a new preparer who took four hours on a return was, in those four hours, building the understanding that made the output meaningful. AI collapses the time but not necessarily the understanding. A junior can now generate a review-ready work product in minutes without grasping why it’s right, what assumptions it embeds, or where it might be wrong. The output looks finished. The accountant behind it may be hollow.

The research names this directly: AI outputs must be treated as inputs to the firm’s quality-control system — not replacements for professional judgment. But a junior who never built judgment can’t supervise the AI’s output — they rubber-stamp it. The faster the tool, the more dangerous the understanding gap, because the errors flow downstream at machine speed with no human check that actually works.

The goal of training in an AI firm isn’t faster output — AI already provides that. It’s building the understanding that lets a human catch the AI when it’s confidently wrong.

What Firms Must Now Teach Intentionally

Because these capabilities no longer accumulate by accident, firms have to build them on purpose. The intentional curriculum for an AI-era firm:

Notice the through-line: nearly every one of these is a judgment skill, and the AICPA’s prescription for building them is consistent — simulation-based learning, AI role-play, and scenario training replacing passive lectures. You don’t build judgment by watching a webinar. You build it by practicing decisions in realistic scenarios, with feedback — the same way the old apprenticeship built it, but compressed and made deliberate. (See developing advisory skills in accountants for how the judgment layer gets built.)

Why Shadowing Breaks Even Faster in an AI-Enabled Firm

If shadowing was a weak training method before AI, it’s actively dangerous now. Here’s why the old “sit next to a senior and absorb it” approach collapses faster in an AI firm:

1. The work to shadow is disappearing. Shadowing relied on watching someone do the repetitive work. When AI does that work, there’s less to watch — and what’s left is the judgment layer, which is invisible. You can’t shadow someone’s professional skepticism.
2. AI adoption adds inconsistency on top of inconsistency. Shadowing already produced different training depending on who you sat beside. Now layer in that each senior uses AI tools differently, trusts them differently, and supervises them differently — and the new hire absorbs a random blend of habits, some of them bad.
3. Stretched managers can’t model judgment they don’t have time to explain. The whole value of the judgment layer is the reasoning behind a decision. A buried manager pushing AI-assisted work out the door doesn’t narrate that reasoning — so the junior sees the output and the speed, but never the judgment. They learn to operate the tool, not to supervise it.

The result is the worst-case AI outcome: staff who are fast with the tools and weak in the judgment that’s supposed to make those tools safe. (Full breakdown in structured training vs. shadowing.)

The New Training Model: Execution → Judgment → Ownership

If the accidental apprenticeship is gone, what replaces it? A deliberate, staged progression that builds each layer the old model produced by chance — in order, on purpose, with the judgment layer made explicit rather than hoped for.

The power of the staged model is that it makes visible — and teachable — what the old apprenticeship left to chance. Execution before judgment, judgment before ownership, each built deliberately rather than absorbed by accident over a decade of grinding that AI has now compressed away.

How Firm Owners Should Prepare Now

Five concrete moves to build the new model before the development gap shows up as a capability crisis:

1. Audit your workflows for the training you’re about to automate away. Identify the repetitive work AI is removing — and recognize that wherever you remove it, you’ve also removed a training ground you now have to replace deliberately.
2. Define readiness explicitly. In the old model, “ready” was a vague sense after enough reps. Now it must be defined: what can this person execute, judge, and supervise? Write it down.
3. Build assessments that test judgment, not just output. Since AI can produce the output, your assessments must verify the understanding behind it — can the person explain the work, catch a planted error, supervise a flawed AI result?
4. Train judgment through simulation. Follow the profession’s own prescription: scenario-based, simulated practice with feedback, not passive content. This is how judgment gets built when the repetitive work that used to build it is gone.
5. Document escalation and AI-supervision rules. Make explicit what staff must do when AI output looks wrong, when a return is beyond their authority, and how human review of machine work actually functions. Don’t leave it to absorbed habit.

Frequently Asked Questions

How is AI changing CPA training?

AI is removing the repetitive entry-level work — data entry, reconciliations, vouching, basic returns — that historically trained new accountants by accident. As that “accidental apprenticeship” disappears, training must shift to what AI can’t provide: judgment, professional skepticism, client communication, and the ability to supervise AI output. The AICPA’s research calls for simulation-based and scenario-driven learning to replace passive lectures, because judgment can no longer accumulate through repetition the machine now performs. The result is a new, deliberate training model that builds execution, then judgment, then ownership thinking on purpose, rather than hoping these capabilities form over years of grunt work that AI has compressed away.

Will AI replace entry-level accountants?

No, but it is reshaping their role and removing much of the routine work they traditionally did. AI automates repetitive tasks — collecting documents, extracting data, reconciling, producing review-ready work products — but accounting still requires human judgment, professional skepticism, client relationships, and the ability to supervise and challenge AI output, none of which AI can replace. The Bureau of Labor Statistics still projects employment growth for accountants. The real change is that entry-level staff must develop judgment and AI-supervision skills sooner, because the repetitive work that used to slowly build those skills is being automated. Firms that train these capabilities deliberately will have capable juniors; those that don’t will have fast operators who can’t catch the machine’s errors.

How should CPA firms train staff in the age of AI?

Through a deliberate, staged model rather than the old accidental apprenticeship: first build execution (the ability to do clean work by hand and recognize what “right” looks like), then develop judgment (interpretation, professional skepticism, client communication, AI supervision), then grow ownership thinking (the path to partner). The methods that work, per the profession’s own guidance, are simulation-based and scenario-driven practice with feedback, not passive lectures — because judgment is built by practicing decisions, not watching content. Critically, staff must learn to do the underlying work before they supervise AI doing it, because you cannot catch a machine’s error if you’ve never produced the correct version yourself. Assessments should verify understanding and judgment, not just output, since AI can generate the output.

What skills do future CPAs need?

On top of technical knowledge (still the entry ticket), future CPAs need judgment-layer skills: critical thinking and professional skepticism to challenge AI outputs rather than accept them; accounting judgment to recognize what’s reasonable or anomalous; client interpretation and communication to translate numbers into advice; escalation skills to recognize the limits of their authority; and AI supervision to evaluate, challenge, and govern automated work. As the AICPA frames it, AI provides speed and analytical power while accountants provide judgment, context, and integrity. The shift is from executing tasks to judging outputs, which makes critical thinking and the ability to ask the right questions the defining professional skills of the AI era.

How does AI affect accounting firm onboarding?

AI makes structured onboarding more urgent, not less. The old onboarding model relied on giving new hires repetitive work and letting competence accumulate through repetition and review notes. With AI automating that repetitive work, there’s less to learn on and the gap between fast output and real understanding widens — a junior can produce AI-generated work faster than they can understand it, then rubber-stamp the machine’s errors. Effective AI-era onboarding must therefore deliberately build execution understanding first (doing the work by hand to know what right looks like), then judgment and AI-supervision skills, through structured, simulation-based practice with gated assessments that verify understanding rather than just completion. Shadowing breaks down further because the work to shadow is disappearing and AI adoption adds inconsistency.

Why is judgment harder to train in the AI era?

Because judgment was historically built as a byproduct of doing repetitive work thousands of times, and AI is removing that work. The accidental apprenticeship — where pattern recognition and the instinct for when a number is wrong formed slowly through grinding reconciliations and returns — no longer happens automatically when the machine does the grinding. Judgment now has to be built intentionally through simulation, scenario practice, and feedback, compressing what once took years of accidental exposure into deliberate development. The added difficulty is that AI lets staff produce correct-looking output without the understanding behind it, so firms must specifically train and assess the judgment to supervise AI, rather than assuming it develops on its own as it once did.

The Bottom Line

For a century, the accounting profession trained its people by accident — hand them the repetitive work, and judgment would slowly form as a byproduct. AI just broke that machine. The repetitive work is being automated, which is wonderful for production and quietly catastrophic for development, because the work that’s disappearing was the work people learned on.

The profession’s own bodies have named the shift: training must move to judgment, simulation, AI supervision, and human skills — taught deliberately, because they no longer accumulate on their own. The firms that thrive won’t be the ones with the most AI tools; tools are becoming universal. They’ll be the ones whose people can supervise those tools — who can look at a fast, clean, AI-generated output and know whether it’s right, what it means, and what to do about it. That capability is built through a deliberate progression: execution, then judgment, then ownership.

AI didn’t eliminate the need to train accountants. It eliminated the accidental way we always did it — and handed every firm the same choice: build judgment on purpose, or end up with a generation that can run the tools but can’t catch them when they’re wrong.

Want to build the new training model — execution, judgment, and ownership — on purpose?

Book Your Free 10-Minute Structural Alignment Review →

In ten minutes we’ll show you how structured development replaces the accidental apprenticeship AI removed — building execution, judgment, and ownership through simulation and gated practice inside the software your firm already uses, so your people can supervise AI instead of rubber-stamping it.

To your firm’s capacity,

Vincent Howard, CPA
Managing Partner, Howard, Howard and Hodges
SkillAbility for Accounting Firms

© 2026 SkillAbility for Accounting Firms. 45-Day Out-of-Pocket Performance Guarantee applies to qualifying onboarding engagements. Contact us for full terms.