AI

If polished output becomes as easy to produce as heating up dinner, what will clients still pay your accounting firm for?

David Leary calls this AI’s “microwave moment.” Anyone can get useful results without the setup that once required technical skill. But easier tools don’t guarantee better service. They can also send generic replies or change a database without asking.

In Episode 507 of The Accounting Podcast, David and co-host Blake Oliver explore what firms can offer when competitors (and clients) have similar tools. Their discussion points toward personal attention, accountable automation, professional judgment, and relationships beyond the deliverable.

When Useful AI Requires Less Setup

David tested Meta’s AI, Muse, while researching bar events before a Bills game in Los Angeles. It produced a polished, web-page-style schedule. Blake noted that Claude could create something similar if asked for an artifact. David’s point was that he hadn’t needed to ask.

“I didn’t have to create projects,” David said. “I didn’t have to configure background agents working. I didn’t have to ask for an artifact. I didn’t even know I wanted one.”

But Blake questioned the economics. Generating those results uses tokens, which cost money. David noted that basic use is free, with paid subscriptions available to heavier users, but both remained skeptical of Meta’s privacy promises.

Easier, faster work also raises a pricing question. Blake cited PwC partner Jeanelle Johnson, who said clients now expect some projects that took eight to ten weeks to take about four.

“If the time shrinks, what happens to the bill?” Blake asked. He expects consulting fees to fall unless firms deliver more value.

That pressure also affects how clients reach the firm.

Your Client’s AI May Become the Front Door

David asked Muse to book a haircut. It couldn’t because his barber only takes phone calls. He warned that firms without online booking may lose business when clients ask AI assistants to schedule appointments.

Blake imagined a firm with its own MCP server, which is a connection that lets AI assistants work with its systems. Clients could submit documents through their assistant. “No more portals, no more emails.” David pictured an agent finding tax documents in Google Drive, sending them to the accountant, booking an appointment, and paying.

These are future possibilities rather than a current firm setup. David also said he hadn’t connected Muse to his email or calendar because of privacy concerns.

The hosts disagreed about the platform risk. Expensify’s availability in Claude and Xero’s connections with ChatGPT and Claude worry David. He feels like vendors are “training your users to not open your app.” But Blake sees integration as necessary: “If they don’t, somebody else will do it.”

For Blake, AI assistants are becoming the main interface, like an operating system. But a convenient interface can’t replace personal attention.

Better Presentation Is Not Personal Service

After blood work, David received a generic cholesterol message recommending 30 to 60 minutes of daily exercise. His doctor already knows he exercises because they attend Orangetheory classes together. David suspects AI wrote the note, though he couldn’t confirm it.

His concern was that busy accountants would send similarly pointless replies. “Why did I hire this person just to send me a templated email?” Blake asked, capturing the client’s likely reaction.

Presentation still matters. A video Blake shared showed how companies once treated annual reports like carefully designed publications. He recommends firms make reports attractive enough that clients want to open them. David emphasized the experience, not simply making reports fun to look at.

That care should also extend to what AI can change behind the scenes.

Give Agents Limits and Develop Human Reviewers

Blake’s AI, acting through his administrator access, added a “pending” status to the hosts’ advertising database. David fixed it quickly, but said an unnoticed change could have caused calculation errors or overpayments. He reduced Blake’s access afterward.

Blake called it “a good lesson in internal controls.” For his own month-end close, he gives his AI agent a separate Xero login without administrator rights. Its actions appear separately in the audit trail. David agreed agents should have their own inboxes and logins so firms can track activity and control access.

Permissions address one problem. Developing judgment is harder.

Blake described PCAOB chair Jim Logothetis’s distinction between audit’s “science” (transaction work and procedures) and its “art” (skepticism, judgment, and industry knowledge). David wasn’t convinced, suggesting AI is good at those things too.

Blake pictured managers, directors, and partners supervising agents. When David asked where future reviewers would learn judgment, Blake answered, “By doing the work.” He then acknowledged the problem: “How do you make more reviewers if you don’t have preparers, and nobody’s figured that out yet.”

This was a forecast, but both hosts said they haven’t seen meaningful accounting job losses from AI. Blake argued that talent shortages could make AI an opportunity to do more with fewer people.

Even capable teams, though, need a reason for clients to choose them.

Build Something Beyond the Deliverable

Blake discussed Giles Pearson’s Accounting Today op-ed comparing AI in accounting with Spotify in music. Streaming gave musicians global distribution, but competitors gained the same access. Musicians had to build identities, audiences, and experiences beyond recordings.

Pearson warns adopting AI isn’t a strategy by itself. Early adopters may gain an edge, but that advantage fades as tools become widely available.

Blake endorsed Pearson’s advice to build community through seminars, roundtables, and client events. “What musicians today sell is the relationship,” he said. David added: “The experience.”

For firms, the challenge is to pair easier access and faster work with controls, judgment, and service clients recognize as personal.

Listen to Blake Oliver and David Leary discuss these trade-offs in Episode 507 of The Accounting Podcast.

A cluster of Polymarket accounts won 41 of 42 bets on whether KPMG-audited companies would beat earnings estimates, a 98% hit rate. Meanwhile, investors put more than $13 million into tax startup ComplYant, whose founder allegedly claimed to be a CPA. Both stories point to the same failure: trust without verification.

On Episode 505 of The Accounting Podcast, Blake Oliver and David Leary discuss the federal investigation involving the Polymarket bets, fraud charges against ComplYant founder Shiloh Luckey, AI-powered scams, accounting automation, and pressure on tax firms to improve pricing and client service.

Trust is the common thread. As AI takes over more routine work, accountants must prove their value through judgment, skepticism, verification, and a better client experience.

When no one verifies

The Wall Street Journal reported that a mystery trader or group made about $22,000 from bets involving 18 KPMG-audited companies, including Wells Fargo, Home Depot, and DoorDash. Blockchain forensics firm Bubblemaps linked 19 accounts through flows of digital funds, suggesting that one person or team may have controlled them. The accounts won 41 of 42 KPMG-related bets and 82% of bets involving companies audited by other firms.

The pattern isn’t proof of insider trading, and the Journal couldn’t confirm that the accounts belonged to a KPMG employee. Still, the Journal has reported that federal prosecutors are weighing charges against a KPMG employee suspected of betting on client earnings. “How could you know that these companies are going to beat their earnings or not unless you had some kind of information ahead of time?” Blake asked. KPMG said it has zero tolerance for trading on nonpublic client information.

The ComplYant case involves information that was publicly available but apparently went unchecked. Prosecutors charged Luckey with nine counts of securities fraud, three counts of wire fraud, one count of bank fraud, and two counts of money laundering. They allege that she falsely claimed to be a licensed CPA and inflated ComplYant’s revenue, subscriptions, and cash reserves while raising more than $13 million. The indictment also alleges personal spending and a $1.5 million check-kiting scheme. These are allegations, not proven facts. Since we recorded, Luckey has pleaded not guilty.

CPA licenses are publicly searchable. As David noted, a large funding announcement doesn’t prove investors performed careful due diligence. In both stories, serious problems emerged after people assumed someone else verified the facts.

AI raises the stakes for fraud prevention

That need for verification is more urgent as AI makes scams cheaper to run. The hosts cited reporting that AI makes fraud 4.5 times more profitable. A scheme that once required a room full of callers can now operate with a few automated agents on a server.

David pointed to Revolut, a neobank that reportedly released customer information after receiving a social-engineering request from a legitimate government email domain. Revolut’s own guidance tells customers to confirm such requests through official contact information. The same standard should apply inside companies: a familiar domain or convincing voice isn’t enough to establish identity.

Technology can strengthen the defense. KPMG bought a stake in Reality Defender, which detects fake voice and video in real time. Meanwhile, Polymarket’s public blockchain records helped outside analysts identify the suspicious betting pattern. But for technology to support trust, firms need to build it into effective controls.

Automation changes how accountants develop judgment

That same shift creates a training problem. KPMG’s TaxSim, developed with Centaurian AI, uses simulations to help entry-level accountants gain experience. Blake saw value in replacing repetitive administrative work with focused practice. David was more skeptical. If real work disappears, how much simulation replaces it? Blake asked who pays for that training. Firms can’t simply automate junior tasks and expect professional judgment to develop on its own.

KPMG is also working with OpenAI on conversational access to systems like ERP and CRM platforms. That could spare employees from difficult software interfaces, but it doesn’t remove responsibility for understanding the underlying data and controls.

Xero faced the same tension after sponsoring an influencer post that claimed connecting Xero to Claude replaced the management accounts she used to pay an accountant £800 a month to prepare. Accountants reacted angrily, and Xero removed the post and apologized. Its leaders said AI should handle routine work while accountants focus on expert judgment and advice. That distinction only matters if clients can clearly see the value of that judgment.

The bigger threat is a poor client experience

Blake and David tested AI on their own business returns. Blake used Claude Cowork with TaxAct for S corporation and partnership returns. David combined TurboTax, ChatGPT, and QuickBooks reports, asking the AI what to do at each step.

David valued the immediate back-and-forth. When he supplied an accrual-basis report instead of a cash-basis report, the AI flagged it. He didn’t have to wait days for another email or manage a long portal exchange.

Neither host claimed AI was error-free. They discussed mistakes made by professionals and tax software, including a return prepared on the wrong accounting basis and Blake’s mother being incorrectly marked as blind. David cited Christine Gervais’s advice to treat every AI answer like work from a first-year hire and verify it against the tax code. Human review still matters, but the possibility of an AI error doesn’t excuse a slow, burdensome process.

David summarized the problem with a mechanic analogy. When clients pay a professional, they shouldn’t have to manage the work themselves. “It’s not AI,” he said. “The threat is the experience for the client.”

The same pressure applies to billing. After IRS guidance warned that billing clients for time AI saved could be an unconscionable fee, Blake argued that hourly billing was the real issue. Firms should charge for expertise and responsibility, not hours that technology eliminates.

Closing the trust gap

These stories ask the same question in different ways: Can accounting professionals still deliver credible verification and judgment? AI makes fraud easier, but it also improves detection and service. Firms can succeed by combining strong controls with careful review, clear pricing, and processes that respect the client’s time.

Listen to Episode 505 of The Accounting Podcast for Blake and David’s full discussion.

In a survey covered by CFO.com, 55% of respondents earning $150,000 to $200,000 a year support replacing CEOs, CFOs and other senior executives with AI. But Blake Oliver points out someone still has to read the analysis, correct it and decide what happens next.

On Episode 508 of The Accounting Podcast, Blake and co-host David Leary connect that survey to pressure on hourly billing and AI audits with one question: who will stand behind the work?

Wanting an AI Boss Doesn’t Prove AI Can Lead

In the Watch Chest survey covered by CFO.com, support for replacing senior executives with AI varied sharply. Forty-five percent of men favored it, compared with 25% of women. The $150,000–$200,000 income group showed the strongest support, at 55%.

David offered a theory about that income split. Higher earners may work closer to executives and feel their teams do the real work. Blake added that middle managers can be “at the mercy” of bad leaders and then pass unreasonable demands to their staff. Neither explanation was a survey finding.

David also distinguished replacing a CEO from automating the work of staff members who prepare slide decks and carry out the CEO’s plans. Blake suggested project management could be a target.

The survey doesn’t answer whether respondents trust AI to run a company. Are they just frustrated with the people running it? Those are two very different questions.

Automation Shifts Work Into Review and Direction

Blake sees AI absorbing finance work, including forecasting, monitoring, month-end close workflows, strategy support and merger-and-acquisition analysis. But can it replace top executives now?

“Definitely, no, not at this point,” he said. “You have to really monitor it, and you have to correct it.” If you let it operate on its own, “you get terrible results at this point.”

Blake often has 20 different AI chats going at once. He reads the output, gives new instructions and moves between tasks while waiting for responses. David uses several AI agents and described the same strain. More gets done, but constant context switching makes the work exhausting.

Blake prefers directing AI to sitting through meetings and delegating. Still, he doesn’t think he can sustain more than six solid hours of that work a day. That is his personal experience, not a universal limit.

“You’re doing heavier reps with your brain,” he said.

The work has changed, but it hasn’t disappeared. Production speeds up, but review is still essential, leaving firm leaders to figure out what clients should pay for.

The Billable Hour Faces Pressure From Both Sides

Blake sees large law firms as a warning for accounting. Citing New York Times DealBook reporting, he described Ropes & Gray’s work on a system designed to produce detailed due-diligence reports in hours rather than weeks.

Clients want to know what those tools mean for their bills. Some use AI agents to flag invoice entries where automation might have reduced the hours, then challenge charges or demand discounts. In-house legal teams often request 20% to 30% reductions across the board, according to the reporting.

Those requests aren’t proof of equivalent savings. Law firms argue that AI hasn’t saved as much time as clients assume, but allows deeper analysis. That echoes the hosts’ experience. More output doesn’t always mean less effort.

Meanwhile, competitors built around AI from the start offer upfront pricing. Traditional firms face a bind: using AI can reduce billable hours, while avoiding it lets competitors undercut them.

Blake cited a Citi survey of 55 large law firms. Of those, 35% expected to modify their billable-hour models within a year because of AI, rising to two-thirds by 2035. Changing the model doesn’t necessarily mean abandoning hourly billing.

The pressure hasn’t stopped growth. Blake cited Wells Fargo figures showing average large-firm revenue growth above 12% in the first half of 2026, while standard billing rates rose by double digits.

David noted that firms may raise rates to offset fewer hours. Blake argued that this approach eventually pushes clients toward fixed-fee alternatives. “I think accounting will follow,” he said.

AI Audits Need More Than an Audit Label

Blake reported that Illinois became the first state to require independent audits of AI developers’ safety claims. Starting in 2028, major developers must hire an outside party each year to check whether they followed their published safety plans.

But checking compliance with a developer’s own plan doesn’t guarantee its AI is safe.

According to Blake, audits must follow generally accepted auditing standards, but the law doesn’t require auditors to be CPAs. It also doesn’t create a certification, registry, accrediting board or disciplinary body for AI auditors.

“This would be perfect work for CPAs,” he said. “They just didn’t require it.”

David raised a comparison with crypto companies’ reports, which Blake described as “the non-audits they called audits.” Blake worried that reassuring reports could reach the public without enough safeguards for competence, ethics and independence. “The whole point is trust,” he said.

David suggested that the AICPA could respond publicly on behalf of the profession.

The opportunity isn’t simply to produce more reports, but to provide oversight people can trust. AI can speed up work, but the hosts’ discussion shows why review, judgment and responsibility still matter. Accountants can’t take their advantage for granted.

Listen to Episode 508 of The Accounting Podcast for Blake and David’s full discussion of AI leadership, billing pressure and who should audit AI developers.