Accounting firms are racing toward a version of artificial intelligence that does more than answer questions. CPA Practice Advisor recently described Ramp Stack as an AI operating system for firms that can automate parts of the monthly close, including transaction coding, reconciliation, and journal entries, with decisions designed to remain reviewable and auditable.
That is exactly the kind of technology firms should adopt. It also changes how leaders need to think about the first years of an accounting career.
Stanford’s August 12 payroll update found employment among U.S. workers ages 22-25 in highly AI-exposed occupations about 19% below where it would be if it had kept pace with similarly aged workers in less-exposed occupations. The comparable gap was 15% in the July 2025 data vintage and 19% by June 2026. The adjustment appears mainly through reduced hiring rather than elevated separations, and experienced workers show no comparable gap.
For accounting leaders, the risk is easy to see. The routine work firms most want to automate is also the work that traditionally gave a junior accountant thousands of small repetitions. A new hire learned how a client’s records fit together by touching reconciliations, tracing source documents, cleaning schedules, preparing workpapers, and watching a reviewer challenge assumptions. Those tasks produced billable output and built pattern recognition at the same time.
Firms should preserve the learning function that routine work once carried without preserving manual drudgery.
The practical tool is an apprenticeship ledger. For every workflow that AI compresses, record four things: what routine preparation disappeared, what judgment task replaces it for the junior employee, who reviews that judgment, and what evidence shows the employee can handle the task independently.
Consider bank reconciliation. AI can match transactions, surface anomalies, and draft explanations. The junior accountant should spend less time checking obvious matches and more time investigating the unmatched items, deciding which evidence resolves the discrepancy and explaining the conclusion to a reviewer. In tax, AI can organize documents and draft issue summaries. The junior should test whether the facts support the treatment, identify missing information, and prepare the client question. In audit, AI can assemble testing support. The junior should trace exceptions, assess the strength of evidence, and explain why an issue deserves escalation.
This shift also changes the senior role. Coaching time has to become a capacity item, not an invisible extra. A manager who saves six hours through AI should not automatically surrender all six hours to a utilization target. Some of that capacity should fund review conversations, live case walkthroughs, and feedback on junior decisions. Otherwise the firm gets faster current output by weakening its future bench.
The ledger should sit beside the productivity dashboard. Track hours saved, realization, and turnaround time. Add time to independent competence, review reversals, the number of judgment categories a junior has handled, and the share of work in which the junior explains a recommendation rather than merely passing along AI output.
That measurement changes incentives. A partner evaluating an AI project can ask two questions at once: Did this workflow become more efficient? Did it produce a more capable accountant sooner?
CPA Practice Advisor has also reported firms scaling AI training across thousands of professionals and redesigning workflows around higher-value work. The next step is to make skill transfer as explicit as tool deployment.
Accounting has always depended on supervised judgment. AI gives firms a chance to reach that judgment faster. The firms that keep an apprenticeship ledger will know whether their automation strategy is building the next generation of trusted professionals or quietly consuming the training pipeline that made today’s partners possible.
ABOUT THE AUTHOR:
Gleb Tsipursky, PhD, is a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook
Photo credit: Who is Danny/Freepik
Sign in to get access to this free resource, and all of our whitepapers and reports.
Download this content today!
Register Now Already registered? Click here to Log In
RT September 3 2026 at 7:39 am
The two questions you end on - did this workflow get more efficient, and did it produce a more capable person sooner - are worth stealing outright. We run a job board for traveling skilled trades and we use AI on resumes and candidate summaries, and we ran into your problem from the other side: the machine is good at the obvious matches and useless on the unmatched items, which is the only part that teaches anybody anything. What we settled on is close to your ledger - AI drafts, a person owns the judgment, and every number has to trace back to the table it came from. The trades have been running this experiment for years without any AI in it, since the helper who learned the job by fetching and cleaning up got thinned out by prefab. This is where we let AI draft and where a person still signs off, if it is any use as a comparison.