I am a CPA in Houston. I spent roughly nine years in audit and corporate accounting — Big Four first, then controllership and technical accounting roles across energy companies — before I started writing about what happens to that work when software can do parts of it.
These are working papers rather than finished ones. They are written for practitioners, for standard setters, and for the people teaching the next cohort of auditors, and across them they argue a fairly specific position: that agentic AI can run most of the evidence-gathering layer of an audit and none of the judgment layer, and that the interesting engineering problem is the interface between the two.
I publish here rather than submitting to a journal because the standards these papers are arguing about are being drafted right now. Comment periods close faster than peer review opens.
Nothing here is neutral, and none of it is settled. If you think a paper is wrong — particularly the argument that full-population coverage using model-based extraction is not automatically safer than a well-designed sample — I would rather hear it than not.