Audit firms have heard the AI pitch before: faster testing, less manual work. What's harder to pin down is how close that future actually is, and what has to change first to get there. That's the question behind every stalled pilot and every partner meeting where someone asks what's actually different this time.

RobotX co-founder and CEO Robert Hyde joined Björn Roskott, data and analytics partner at Newtone, on stage at PKF's Global Gathering in November 2026 for a session called "AI in the Audit: Empowering the Network."

1. Auditors and developers are equally exposed to AI. Developers are experiencing it first.

The comparison comes from Robert's own background: an auditor who now runs a software company and watches AI reshape how developers work. He sees the same pattern coming for audit.

"We both work with rules and structure," he said. "We both spend a large part of our day checking whether something is right. And we both document everything we do. Which happens to be exactly the kind of work AI is very good at. That's the inconvenient truth."

He translated a framework from engineer Bassim Eledath into audit terms: at level one, AI drafts and the auditor still does the work; in the middle levels, the auditor delegates whole procedures; at the top, procedures run inside a harness of checks the auditor reviews and signs off on.

He placed RobotX well up that curve. "Two years ago, developers wrote code. Today they decide what's needed and judge the result. The profession didn't disappear. The production did. The same will happen in audit, and it won't take ten years."

Audit keeps going, but the job has shifted from execution to judgment. What's a firm doing today to get auditors there? At Newtone, it runs on three levels: generic AI and agents for search and analysis, data analytics across full populations to flag risk and exceptions, and RobotX for running and documenting the substantive testing.

2. The barrier to scaling AI lies in ownership.

A second poll asked firms their biggest barrier to scaling AI and automation. Leadership and ownership came out on top, ahead of data quality, integration and skills.

What surprised them was what came near the bottom: trust and governance. Governance reads as solvable. Ownership is the problem nobody wants to own.

The practical take from both: a tool needs a clear owner, and it has to sit inside the process, not next to it. "Most firms run a pilot and it works. But it will not scale if it's placed next to the process instead of inside it."

Auditors don't want another login, another tool, season after season. "The reaction I hear most," Robert said, "is: please, not another tool."

His advice for any pilot: don't just ask whether it works. Ask how many manual steps still sit around it.

Ownership and integration are separate failure points. Ownership decides who's accountable for adoption; integration decides whether a working tool actually gets used day to day. A firm can fix one and still stall on the other. The firms that scale ask where a tool sits and what it connects to before they ask which tool to buy.

3. The time is saved where the steps connect.

The real time leaks out between the layers, in the handovers: a client sends data, someone exports and uploads it, and waits; a result comes back, gets downloaded, and the team waits again. "Hours to days of elapsed time," he said. "Most of it is waiting, not working."

The model he sketched has four layers: methodology drives the tests, client data is connected, execution runs the tests and gathers evidence, and the result lands in the audit file as a working paper. "A delivered request starts the work. The result lands in the file," he said. "Optimize each layer. Then connect them. That's where the more time is saved."

Looking at a full population doesn't automatically make an audit more efficient. It usually surfaces more exceptions. The gain comes once chasing those exceptions, gathering evidence, and documenting the result runs through the same connected layers.

4. The harness is where the audit lives.

Asked what a technology partner should actually look for in a tool, Robert pointed toward whether it can grow with the firm into a more connected audit stack.

"The model is just the engine," he said. "It doesn't matter if it's Anthropic or OpenAI. What matters is the harness around it, the framework that makes sure that intelligence does well."

Think of a Formula 1 car: the model is the engine, but the harness wrapped around it is what lets that same intelligence run reliably on financial statements, checking internal consistency and numerical bias. "That's where the truth is," he said.

Björn added the rule that follows: keep the harness, swap the model underneath when a better one comes along. "The best model of today is definitely not the best model of tomorrow," he said. "Keep your technology open."

The technology does not run on its own. Every procedure still hands off to a person for exceptions, source validation and judgment, and both speakers agreed that it stays that way. The harness exists so trust stays with the auditors.

5. Closing the gap starts with the data.

Most firms are still behind where they'd like to be. Their advice on closing that gap: start experimenting with generative AI now, so people build the habit, and build the underlying data capability in parallel, because that's what the next wave of audit technology will depend on. Buying the best tools on top of unreliable data still produces an unreliable result, whatever the tool claims to do.

Newtone is an example of why the groundwork on data needs to start well before AI enters the picture. Years before any AI sat on top of it, Björn's team built a background-check agent in-house: SQL queries and manual cross-checking. Member firms are independent, and what one firm develops and funds stays its own. What travels across a network is the knowledge: the lessons learned, and the dead ends the next firm can avoid. Shared openly, that experience means nobody starts from a blank page.

That's what "Empowering the Network" was actually asking: whether member firms keep solving the same data problem in isolation, or share what they learn doing it, the missteps and the parts that worked.

Everything else, the ownership, the connected layers, the harness, is just what makes that technology work. The hour ended on Robert's line: "Whatever the future brings, the profession of tomorrow is about relationships, not numbers. But to get there, you have to start with technology today."

Get started with RobotX →