Skip to Content

2027: Year of the Meat Proxies

Blogs Richard Penfil Sep 17, 2026

Ahead of Richard Penfil’s AI presentation at the 2026 Ignite Conference, Nov. 2-3, he conjures a not-so-distant future, in which auditors are sidelined by AI.

The chief financial officer put next year’s audit plan on the screen. It covered more ground with fewer people. He pointed to our completion figures. “You managed it this year,” he said.

Eighteen months earlier, the results had been worth celebrating. AI (artificial intelligence) agents had collected evidence, reconciled files, documented tests, and drafted findings. Auditors spent less time chasing attachments. Reports arrived sooner. The CFO asked why a department producing more work needed the same head count. So, we stopped replacing people who left.

The next budget assumed the savings would continue. We expanded coverage again. Experienced auditors inherited more engagements, while AI agents handled the preparation. Reviews remained mandatory, but our metrics only focused on completion dates and reports issued. We never established how much work a human reviewer could meaningfully handle.

When management asked for another round of savings, junior auditors lost their jobs first. We canceled the graduate intake too. I helped one departing auditor rewrite her resume. We had told her that routine testing was how she would learn to lead an audit. She was still learning when we decided we no longer needed her.

Meanwhile, the agents improved. They began identifying the weaknesses we would have raised in planning meetings. They weighed conflicting evidence, suggested further testing, and drafted conclusions that experienced auditors began accepting verbatim. Helen, our strongest audit manager, spent an afternoon comparing an agent’s assessment with her own. She preferred parts of its argument. So, we gave it more responsibility.

Soon Helen was only reviewing the finished assessments instead of evidence. The agent had selected the questions, pursued the inconsistencies, and decided how strongly to state the finding. She could reopen any of it. But while she reconsidered one conclusion, more AI-completed work piled up.

Management saw another opportunity. If the agent could now do work previously assigned to a manager, surely each manager could supervise more engagements.

Our mornings disappeared into approvals. Investigating and deep diving became something we had to justify. A careful reviewer fell behind. A quick one cleared the list and received another assignment.

The agents were often right. That made it easier to accept the next recommendation. It also made spending an afternoon challenging one look increasingly wasteful.

“Human judgment remains essential,” we told the audit committee. We put a named reviewer on every report.

Someone on the team started calling us meat proxies. We supplied the signatures, attended the meetings, and took the questions when a conclusion was challenged. The agents had already done most of the work we had once enjoyed discussing.

Helen resigned that autumn. “I am no longer needed here,” she told me.

She took a role where she could contribute again. Another experienced auditor moved elsewhere in the business. We posted positions for replacements. Recruiters found qualified people, but several declined interviews after learning how much of the job involved reviewing AI output.

The remaining juniors could not take Helen’s place. They still needed supervision, and the managers who remained struggled to find time to teach. We needed an experienced reviewer immediately. Canceling recruitment also meant fewer people would be developing those skills for the years ahead.

Our strongest candidate spent an afternoon with the team. She watched the recommendations arrive and asked how much time she would have to work on anything other than reviewing. She withdrew the next morning.

At the budget meeting, the CFO asked whether another agent could review the output. We closed the vacancy.

A week later, an employee caught me after a meeting. “Please don’t use my name,” she said. “We’re told to record complaints as inquiries. Nobody wants bad numbers.” I promised to follow up. Then I postponed the meeting to finish the reports awaiting my approval.

By the time a whistleblower report reached the board, customers had been overcharged for months. Our agent had reported falling complaint rates from the altered records. One employee had been disciplined for questioning the figures.

Customer compensation and investigation costs exceeded the savings from cutting our audit team. The CEO asked why nobody had warned him. I found the postponed meeting in my calendar.

Richard Penfil will discuss how to avoid this future at his presentation, All Hands on Keyboard: Techniques from the Bleeding Edge, at the 2026 Ignite Conference, Nov. 2-3, in Washington, D.C. and virtually.

The views and opinions expressed in this blog are those of the author and do not necessarily reflect the official policy or position of The Institute of Internal Auditors (The IIA). The IIA does not guarantee the accuracy or originality of the content, nor should it be considered professional advice or authoritative guidance. The content is provided for informational purposes only.

Richard Penfil

Richard Penfil is a data science manager at PayPal and is based in Austin, Texas.