Members across firm sizes and specialties came together to explore what AI is actually doing to—and in—the accounting profession. Moderated by Ron Baker and Ed Kless, hosts of The Soul of Enterprise, the conversation provided a clear picture of where the profession stands and where it needs to go.

The Experiment That Started the Conversation

Much of the early discussion centered on a recent book documenting what happened when someone set up an AI-powered accounting firm using multiple AI agents, each assigned a specific role. The output was reviewed by a credentialed third party, who described it as impressive and identified issues that human reviewers had missed. A highlighted finding was that the human directing the operation was consistently flagged by the AI agents themselves as the primary bottleneck, not because the human wasn’t valuable, but because the system’s capacity to produce work exceeded what one person could direct and review.

The group didn’t treat this as a warning or a promise. It was framed as a grounded, honest look at what AI can do today, and a signal that the profession needs to redefine what human involvement looks like rather than defend the status quo.

What Humans Bring That AI Can’t

A recurring theme was the irreplaceable nature of human judgment, relationships and accountability. Several perspectives landed and resonated with the group:

Clients don’t stay with professionals because of unique technical skills. They stay because of the trust built over time through real human interaction. AI can’t take a client to lunch or sit with someone through a hard moment.

The medical analogy came up more than once. Most people want AI to help their doctor diagnose a disease. But no one wants a bot to deliver a serious diagnosis. The way difficult information is delivered matters enormously, and that’s entirely a human responsibility.

Legal and professional accountability can’t be delegated to a tool. No one can tell the IRS that the bot made a mistake.

Clients are also arriving at appointments with more AI-generated knowledge than they used to. The role of the professional is evolving from primary knowledge source to trusted interpreter and guide, which is a higher-value role, but one that requires a different mindset and skill set.

The Billing Model Conversation

This was one of the most energized parts of the discussion. The core issue is simple: if AI compresses a 10-hour task to 15 minutes, the hourly billing model doesn’t hold.

Some firms are already choosing to bill for the value delivered, not the time spent, even when AI does the work faster. Others haven’t figured out how to have that conversation with clients yet.

A real-world example was shared involving a large firm demanding a significant fee reduction from an outside firm after learning AI was being used in the work. The outside firm complied. The group saw this as a cautionary signal about what happens when the value conversation doesn’t happen proactively.

The consensus was that value-based and subscription pricing aren’t just theoretical anymore. The hourly model has been under real pressure, and firms that don’t get ahead of it will be in a reactive position.

Specialized Tools vs. General-Purpose AI

Members had a wide-ranging conversation about the tradeoffs between specialized AI tools built for accounting and general-purpose platforms. Different firms have had different experiences, and there was genuine debate about which factors matter most when evaluating options, including workflow fit, flexibility, cost and how each tool type handles customization and experimentation.

The group recognized that there isn't a one-size-fits-all answer and that the right choice depends heavily on a firm’s specific context and its overall approach to AI adoption.

Accuracy, Over-Reliance and Professional Judgment

A few themes surfaced that centered on concerns about AI making mistakes.

AI speaks with confidence even when it’s wrong, which can mislead users who associate confident delivery with correctness. That’s a useful reminder for the profession that critical thinking matters regardless of the source.

The risk of de-skilling was raised, drawing a parallel to how calculators affected math skills. As professionals rely more on AI, there’s a question about whether they’re maintaining the foundational knowledge needed to catch what AI gets wrong.

At the same time, the standard isn't perfect. Humans aren’t 100 percent accurate. AI only needs to be better than current human performance, and on many routine tasks, it is. The goal is human oversight of AI output, not the replacement of human judgment.

AI as a Learning Tool

A thread that drew genuine enthusiasm was AI as a personalized tutor. Examples were shared of students and professionals uploading textbooks or course materials and using AI to generate practice questions, receive explanations for wrong answers and progress at their own pace. The phrase “a tutor with infinite patience" came up. This is already being used in academic settings with positive results.

Using AI to Differentiate Your Firm

One of the more forward-looking conversations was about whether and how firms should use AI adoption as a market differentiator. The group agreed that the differentiator isn’t claiming you use AI. It’s being specific about how you use it and what tangible benefit it delivers. Accuracy, speed, cost efficiency and better outcomes are things clients can evaluate.

As AI becomes standard across the profession, the differentiator will shift from “we use AI” to “here is specifically how we use AI to deliver better outcomes for you.” Firms that develop clear, credible AI use cases will be ahead of that curve.

CalCPA Conversation Circles are free, member-only experiences designed to foster conversation and connection. Be sure to register for our next one on July 8, when we’ll rethink what innovation means beyond technology; explore how language, mindset and processes shape outcomes; and provide new perspectives you can apply in your organization.