UpTrajectory Review

Shannon Smith's piece in CPA Practice Advisor tackles the question sitting on every audit partner's desk this busy season: which parts of the audit can artificial intelligence absorb, and which must remain human work. The article frames the divide as a matter of responsibility, not capability. AI can grind through transaction matching, anomaly detection, and the repetitive sampling that consumes junior staff hours. Judgment calls, client context, and the final opinion stay with the auditor. Smith's framing matters because the profession has spent two years debating whether AI will replace auditors, while the sharper question asks which tasks AI should take. That shift from fear to design changes how a firm buys software, trains staff, and prices engagements.

Small firms feel this pressure first. A three-partner shop cannot throw bodies at busy-season volume the way a national firm can, so automation looks like survival. Yet the same firm carries the same professional liability if an algorithm misflags a material misstatement or misses one entirely. Smith's distinction gives smaller operators a usable filter: automate the evidence-gathering, keep the reasoning. A firm that lets AI draft procedures or rank risks without human review trades short-term capacity for long-term exposure. Partners should read the piece as a checklist for vendor demos, not a think-piece. Ask any AI audit tool where it stops and where the human must step in; a vendor that cannot answer clearly is selling liability, not efficiency.

What the article likely underplays is the training gap. Telling a firm to keep judgment in-house assumes the humans still know how to exercise it after years of delegating routine work to software. Junior auditors learn by ticking and tying. Remove the drudgery and you remove the apprenticeship. Smith draws the line correctly, but the profession has not yet solved the pipeline problem her line creates. Firms need a deliberate plan for how staff develop skepticism when the machine handles the first pass. That plan might include rotating juniors through manual review of AI output, or pairing them with partners on judgment-heavy sections. Without such a plan, the firm gains speed now and loses competence later.

The downstream effect reaches clients too. Auditors who automate well can cut fees or redirect hours toward advisory work, which changes what small businesses buy from their accountant. A client who once paid for forty hours of fieldwork may now pay for ten hours of review plus strategy. That shift rewards firms that price on value, and it punishes firms that bill on hours. Smith's question about where AI belongs is also a question about which business models survive. Owners who buy audits should watch how their firm responds; a partner who slashes fees without explaining the new workflow may be cutting scope, not waste.

Watch how the standard-setters respond. The AICPA and PCAOB have signalled scrutiny of AI-assisted audit evidence, and regulators will not wait for consensus. Firms should document every point where a human overrides or accepts an AI output, and they should treat that documentation as the product, not the byproduct. The practical step this week: map your own workflow, mark the tasks that require professional skepticism, and test whether your current software respects that boundary. If it does not, the tool is running your audit, not supporting it. Smith's article gives the framework; the firm must supply the discipline.

“What work should technology take off an auditor’s plate, and what responsibility needs to stay with the auditor?” — CPA Practice Advisor

Takeaway: Automate audit evidence-gathering, keep human judgment, and document every AI override as if regulators will read it.