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

Artificial intelligence is changing accounting faster than most firms can redesign their training.

The immediate benefit is easy to see. AI can collect documents, extract data, categorize transactions, identify discrepancies, draft summaries, assist with research, and accelerate routine preparation. Work that once consumed hours may soon require only minutes.

That creates a major productivity opportunity.

It also creates a development problem that accounting firms cannot afford to ignore.

For decades, new accountants developed professional judgment by doing the work. They reconciled accounts, prepared workpapers, traced transactions, reviewed source documents, researched unusual items, corrected mistakes, and responded to review notes.

That repetition was not always efficient, but it served an important purpose.

It helped young accountants develop pattern recognition. They learned what a normal transaction looked like. They saw how one error affected several accounts. They discovered that client explanations were not always complete. They learned when a number looked unusual, when support was missing, and when an issue needed to be escalated.

In other words, routine work was also an apprenticeship.

As AI takes over more of that work, firms must answer a difficult question:

How will new accountants develop judgment and professional skepticism when the experiences that traditionally built those skills are being automated?

AI Removes Repetition, but It Does Not Remove Responsibility

An AI system can produce a polished result quickly.

That does not mean the result is correct.

The system may rely on incomplete information, misunderstand the client’s facts, apply a rule incorrectly, overlook an exception, or produce a confident explanation that is not supported by the source documents.

A trained accountant must still decide whether the output can be trusted.

That requires more than software knowledge.

The accountant must understand the purpose of the work, the relevant business context, the evidence required, the risks involved, and the conditions that should cause the normal process to stop.

The future accountant may spend less time preparing the first draft. The accountant will spend more time evaluating whether the first draft is accurate, complete, reasonable, and appropriate for the client.

This is why firms need a different approach to AI accounting training. Teaching employees how to operate an AI tool is not enough. Firms must teach them how to question, verify, document, and overrule the tool.

The Disappearing Apprenticeship Problem

Traditional accounting development was often informal.

A new employee completed a task. A senior reviewed it. The work came back with corrections. The employee fixed the file and gradually learned what the reviewer expected.

The model was imperfect. It consumed manager time and often depended too heavily on shadowing. However, it exposed employees to the details behind the final result.

AI may now complete many of those details before the employee sees them.

Consider a junior accountant reviewing an AI-generated reconciliation.

The balance agrees. The report looks professional. The explanation sounds reasonable.

But does the employee understand:

• Where the information originated?
• Whether the source data was complete?
• Why the account needed to be reconciled?
• Which reconciling items are unusual?
• Whether the difference belongs in the current period?
• What additional evidence is needed?
• When the issue should be escalated?

An employee who never learned the underlying process may accept a convincing output without recognizing that it is wrong.

The firm gains speed but loses control.

Professional Skepticism Must Become a Deliberate Skill

Professional skepticism is sometimes described as an attitude.

In practice, it is a repeatable behavior.

A skeptical accountant asks:

1. What evidence supports this result?
2. What information could be missing?
3. What assumption is the conclusion relying on?
4. What would make this answer wrong?
5. Does the result make sense in the context of the client’s business?
6. Is this issue within my authority, or should it be escalated?

These questions should be built into training, workpapers, review procedures, and client discussions.

New accountants should not be taught to distrust every automated result. They should be taught to verify the result in proportion to the risk.

A routine classification may require a quick comparison with the source document and prior treatment.

A material tax position, unusual journal entry, significant estimate, or client recommendation may require deeper research, corroborating evidence, and manager review.

The objective is not suspicion for its own sake.

The objective is disciplined confidence based on evidence.

Firms Need Practice Before Live Client Work

Historically, the first place many accountants practiced a skill was inside a live client engagement.

That approach becomes more dangerous when AI can make incomplete work appear finished.

Firms need controlled practice environments where employees can make mistakes without creating client risk.

A realistic exercise might give the learner:

• Source documents with missing information
• An AI-generated workpaper containing subtle errors
• A prior-year file with a treatment that should not be repeated
• A client explanation that conflicts with the records
• A deadline that requires prioritization
• An issue outside the learner’s authority

The employee should be required to review the work, identify the problems, document the evidence, correct what is appropriate, and escalate the remaining issues.

The assessment should evaluate the reasoning, not just the final number.

Did the employee identify the missing support? Did the person recognize the unusual transaction? Did the employee challenge the AI-generated conclusion? Was the escalation early, clear, and supported by facts?

These are the behaviors that protect clients and firms.

Managers Must Stop Being the Entire Training System

AI does not eliminate the need for managers.

It changes where managers create value.

Managers should spend less time repeatedly explaining routine procedures and more time developing judgment, review quality, communication, and decision-making.

That requires managers to coach differently.

Instead of immediately giving the answer, the manager can ask:

• What did you review?
• What evidence did you find?
• What does not make sense?
• What alternatives did you consider?
• What do you recommend?
• What would cause you to escalate this issue?

These questions make the employee’s reasoning visible.

The manager can then determine whether the gap is technical knowledge, missing context, weak investigation, poor self-review, or a failure to recognize risk.

Correcting the file solves one assignment.

Correcting the reasoning improves future assignments.

Training Must Progress From Execution to Leadership

AI makes leadership development more urgent, not less.

When routine preparation shrinks, firms will need accountants who can review automated output, explain financial results, communicate with clients, coach staff, manage workflows, and make sound decisions earlier in their careers.

That progression should be intentional.

Early-career accountants should learn to:

• Understand the underlying accounting process
• Verify source information
• Recognize exceptions
• Document conclusions
• Escalate uncertainty
• Produce review-ready work

Experienced staff and seniors should learn to:

• Review the work of others
• Identify material risk
• Explain financial and tax consequences
• Coach junior employees
• Lead defined client conversations
• Recommend appropriate next actions

Future managers should learn to:

• Delegate responsibility
• Protect review quality
• Develop people
• Manage deadlines and engagement economics
• Handle difficult client conversations
• Recognize advisory opportunities
• Build capacity through the team

A firm that wants to remain competitive must do more than train employees to use new technology. It must deliberately develop accounting leaders who can apply judgment when the technology reaches its limits.

The Competitive Advantage Will Be Human Capability

AI tools will become more accessible.

The ability to use them will not remain a meaningful differentiator by itself.

The competitive advantage will belong to firms whose people can determine:

• Whether the output is reliable
• What the result means
• Which issue matters most
• What the client should do next
• When another professional should be involved
• How to develop the next person to make the same decision

Firms that treat AI as a replacement for development may create fast but fragile teams.

Firms that redesign development around judgment, skepticism, communication, and leadership can gain the productivity benefits of AI without weakening professional capability.

The old apprenticeship model depended on learning through repetitive production.

That model is disappearing.

Accounting firms now need to build the learning process intentionally.

The goal is not to preserve inefficient work simply because it once trained people.

The goal is to replace accidental learning with structured practice, realistic assessment, focused coaching, and progressive responsibility.

AI can produce the first draft.

Accounting professionals must still determine whether it deserves to become the final answer.

The firms that understand that distinction will not only adopt AI more effectively.

They will develop the reviewers, advisors, managers, and future leaders the profession needs next.