Not Your Grandparents' AI: Inside the End-to-End Agentic Audit

At Catalyst 2026, I demoed the end-to-end agentic audit: Field Agents across planning, testing, reporting, and review, all running on one shared engagement record. Watch the full demo below.

Ask an auditor where an engagement lives and the honest answer is everywhere. Scope is in the engagement letter. Controls are in the client's spreadsheet. The walkthrough is in a notes app, requests are in PBC software, testing is in a workpaper, and the report is in Word.

That's six places, and none of them talk to each other. So teams retype, reformat, and cross-reference by hand, and sometimes ask the client for the same document twice. The judgment and skepticism auditors trained years for gets whatever time is left at the end.

Most AI in audit so far hasn't changed that picture. It adds a seventh place: a tool on one screen where you paste something in, copy something out, and carry the result back to wherever the work actually lives. Useful, but the engagement is still in six pieces.

At Catalyst, I showed what it looks like when the AI agents orchestrate the whole engagement instead.

One record from start to finish

Field Agents work from one shared engagement record. Field Planner reads the engagement letter and risk control matrix, updates controls from walkthrough notes, and drafts requests already linked to the controls they satisfy. In the demo, that was 47 requests covering 133 controls, with single requests linked to multiple controls, the mapping teams usually spend hours on doing manually

Because it's one record, what Field Planner scopes is what Field Auditor tests, and what Field Auditor concludes is what the report pulls in. Nothing gets re-keyed between steps.

Purpose-built for each workflow

Field Auditor took a wire authorization control from the test plan to handoff package. It built the test plan, ran the completeness and accuracy check, pulled a random sample, marked up the evidence, and recorded insufficient support for one approval as a deficiency in the test sheet.

On a financial statement audit, it ran a search for unrecorded liabilities using the firm's own procedure. It flagged a duplicated transaction in the disbursement register, scoped the population, tested the samples, and produced the workpaper.

Field Reviewer applied a reviewer's own criteria to a SOC 2 report and proposed each change in place, so the review happens in one system instead of across several. Field Financials drafted this year's statements from the trial balance in last year's format, with every number linked to its source.

One place to review and run it

Field Board is the kanban-style command center for the Agent Workforce. Drag controls into testing and the agents start; finished work lands in preparer review, ready for the auditor.

But speed only matters if the work holds up, so review is built into how the work moves. Every agent conclusion traces back to the transaction and source document behind it. The auditor pulls up the evidence and overrides any output without leaving the board, and that change carries through to the workpaper and the report.

Practitioner led. Agent executed.

A partner, not a vendor

The AI most firms have seen so far sits next to the work. The end-to-end agentic audit is built directly into it, so auditors spend their time on the part of the engagement that needs them.

I closed the demo at Catalyst with one thought, and it applies here too: as you plan this shift at your firm, look for a partner, not a vendor.