The accounting profession sees AI companies building tools that weren’t meant for accountants but are increasingly doing accountants’ work. On Episode 482 of The Accounting Podcast, hosts Blake Oliver and David Leary discussed Perplexity and Palantir’s moves into core accounting territory while most firms struggle to see any productivity gains from their AI investments.
Trump Accounts Hit 90% Adoption in First Year
Before getting into AI disruption, the hosts discussed a surprising government success story. Around four million children have been signed up for Trump Accounts in the program’s first year. That’s an 85-90% adoption rate among eligible births since January 2025.
“Think about 401(k) plans,” David pointed out. “It’s been decades and they’re only at 35-40% participation. With college savings plans, it’s been 25 years and 25% participation.” The difference is Trump Accounts combine simplicity (a one-page form filed with your tax return) with immediate value (a $1,000 government contribution, plus additional funds from donors like Michael Dell for low-income families).
Blake ran through the potential impact. If families contribute the maximum $5,000 annually, a child could have $271,000 by age 18 based on historical S&P 500 returns. “That’d be pretty nice,” he said. “Turn 18 and you get $271,000. Maybe that would be a down payment on a house someday.”
Perplexity and Palantir Target Core Accounting Work
The real disruption story started with a simple Instagram ad that caught David’s attention. It read “Find every duplicate entry hiding in your QuickBooks.” The advertiser wasn’t a QuickBooks app developer or accounting software company. It was Perplexity, an AI company better known as a search engine competitor.
Clicking through revealed Perplexity’s new QuickBooks “health check” offering. It includes P&L analysis, expense categorization, AR/AP aging reviews, and reconciliation audits. Perplexity’s homepage also features tax-specific prompts, and they’ve officially launched “Computer for Taxes,” an AI agent that drafts full federal tax returns.
“Perplexity is doing what a whole company just launched to do,” David observed, referring to startups like TaxGPT that built entire businesses around AI tax prep.
Meanwhile, the IRS is testing a $1.8 million pilot with Palantir to build an AI system called SNAP for selecting audit targets. The system will analyze both structured and unstructured data to identify potential fraud and noncompliance.
“People have called Palantir the most dangerous company on earth,” David noted. “So that’s who’s going to pick audits now for the IRS. It’s a little bit scary.” The system could potentially cross-reference tax returns with e-commerce storefronts, social media activity, and data from other government agencies where Palantir operates.
Blake’s Test Whether AI Actually Does the Books
Blake decided to test whether these AI tools could handle real accounting work. He asked Claude Cowork to reconcile a brokerage account in Xero that had no bank feed connection.
Without detailed instructions, Cowork created its own task list, asked clarifying questions, then got to work. It parsed the PDF bank statement, converted it to a CSV formatted for Xero’s import requirements, imported it as a bank feed, matched transactions through the Chrome browser, ran the reconciliation report, and exported it as a PDF.
“End to end, no hand-holding,” Blake said. Even better, he had Cowork save the entire process as a reusable “skill” that improves with feedback. “Now I can upload a PDF and say ‘reconcile this account’ and it will do everything as I like it.”
This raises questions for accounting software companies. “What is the point of Jax in Xero then?” David asked. If external AI agents can perform full reconciliation by interacting through a browser, why build internal AI at all?
Blake says Xero built Jax by copying chatbot functionality from ChatGPT. It’s limited to conversation windows with no ability to handle multi-step workflows. “The AI platform companies have simply raced ahead,” he said.
Firms Are Automating the Wrong Things
Despite all this technological capability, 80% of firms report no measurable AI impact on productivity after three years and billions in spending. The problem, according to a CFO.com opinion piece Blake highlighted, is that firms use the wrong measurement framework.
Most organizations use cost accounting, which rewards local efficiency gains, including hours saved here, a task automated there. But they should use throughput accounting, which asks whether AI actually removes the main bottleneck limiting revenue generation.
“You may have theoretically saved all these hours on tax prep, but if all the returns pile up at review, it’s not getting to the client any faster,” Blake explained. “You’re not delivering any additional value.”
Both hosts agreed the real bottleneck in most accounting workflows isn’t doing the work; it’s getting the information needed to do it. Client document collection causes the biggest delays.
Blake sketched a solution: an AI agent that monitors client folders, compares submissions against a requirements list, and automatically follows up on missing or incomplete documents. “AI won’t be tired on day two or day three,” David said. “It can review these things and send the email again tomorrow. And it’s never going to complain.”
The Three-Person Firm of Tomorrow
The hosts painted a picture of accounting’s near future: firms where a few people do the work of ten, supported by AI agents handling staff-level functions.
“Instead of having 30 clients at your firm, you now can have 90,” David said. “That makes a huge difference.”
Blake envisions a new role structure that includes a CPA or tax expert, a dedicated technologist, and an apprentice, plus “half a dozen AI agents doing the staff functions.” He pointed to conversations with Peter McCarroll of Fuel Accountants, who predicts every firm will need a technologist role.
But Blake offered a reality check. “Cowork crashes. Agents are slow. Building a reliable tech stack in a period of this much change is genuinely hard.” Firms don’t need to adopt every new tool. But they do need to understand their workflows well enough to know exactly where to point AI.
General-purpose AI platforms are building accounting capabilities faster than the profession anticipated. They’re not waiting for permission or partnerships. Listen to the full episode for more insight into figuring out where your real bottlenecks are and pointing AI at them, or watch as others (maybe even your clients) use these tools to work around you.
