EY just gave its auditors a staff of software agents. A profession built on professional doubt now has to decide how much of that doubt a machine can carry.
Scale first
EY wired a multi-agent framework into EY Canvas, its single global audit platform. It runs on Microsoft Azure, Foundry and Fabric.
Canvas serves 130,000 assurance professionals on 160,000 engagements across more than 150 countries and territories. It also processes over 1.4 trillion lines of journal entry data a year.
EY says extensive testing and piloting came first. The global rollout now reaches every phase of the audit, and EY expects agents to support all end-to-end audit activities by 2028.
CFOtech ranks it among the broadest AI deployments a professional services network has disclosed in audit work.
Doubt is the product
An audit sells one thing: an independent professional’s refusal to take management’s word for it. Agents can orchestrate tasks, flag risks and pull current accounting guidance. They can’t doubt like the person who signs the report.
EY says it built the approach to keep human judgment and skepticism in place. CEO Janet Truncale calls the audit “human-led, AI-powered.”
Alina Dimitriu, EY Romania’s assurance leader, splits the work the same way. Technology handles the volume and the risk. People keep the responsibility.
The apprenticeship gap
The release skips one problem. Junior auditors learn skepticism by grinding through checks, and agents now absorb much of that grind.
Take away the hours spent tying out ledgers by hand and you take away the training. Firms need another way to build judgment.
EY’s answer is a global training program for every audit and technology risk professional this year, with in-person learning. It’s a start. Nobody knows yet whether it replaces what the grind taught.
Auditing the agents, too
EY also launched assurance services covering AI diagnostics, governance, risk management and controls. Audit and non-audit clients can buy them.
So one firm now audits with agents and audits agents. It also joined the industrial affiliates program at Stanford’s Institute for Human-Centered AI.
One trade outlet expects regulators and standard-setters to scrutinize this kind of adoption and, in time, shape the standards for it.
What to watch
The winning model won’t have the most agents. It will let a human reviewer show what an agent did, what it missed and why the signature still carries weight.
Two things to track: whether EY discloses how it reviews agent work, and how fast the other Big Four firms follow.
Related: Autonomous AI Agents: A 5-Layer Enterprise Framework for 2026
Test this on a real task
The audit is free and takes about ninety seconds. It returns a verdict, a readiness score and the supervision the task would need.
Run the audit