AI reliability in accounting: breaks the same way your staff does
Anthropic surveyed 80,508 AI users. The number one fear isn't job loss — it's unreliability. AI fails the same way your staff does. The difference: corrections become permanent rules.
Peter McCarroll
Founder of Fuel Accountants and The AI Accountant. With over 15 years in practice management and technology consulting for accounting firms, Peter is passionate about helping accountants embrace AI to transform their businesses.
70 articles
QuickBooks AI integration: What the Intuit-Anthropic partnership actually means for your practice
Intuit and Anthropic announced a multi-year partnership putting QuickBooks inside Claude. The MCP server is already public on GitHub — a developer expanded it to 143 tools. But the depth question that's limited every platform integration still applies, and the Enterprise Suite demo isn't what your QBO clients will get.
Weekly AI Roundup: Your clients just got a direct line to Claude
Intuit just connected QuickBooks directly to Claude — your clients can now build AI agents with their own financial data, no accountant required. Plus every AI company is converging into every other AI company, Jeff Bezos is raising $100 billion to rebuild manufacturing with AI, and 81,000 Claude users just confirmed what we published on Wednesday.
Your staff is already putting client data into AI. Here's what that actually means.
98% of accounting firms use AI. Only 21% have a policy governing it. Your staff is already putting client data into tools you haven't vetted — and the line between safe and dangerous isn't where most practice owners think it is. Here's what's actually risky, what's safe right now, and the one thing you should implement this week.
The invisible tax your team pays every day
Before every coaching call, I check with my bookkeeping team to get current on the client's status. That's not preparation — it's a tax. Every CAS practice pays it dozens of times a day: the invisible cost of moving context between people so someone can do the actual work.
The judgment gap: Why accountants are overestimating their moat
Sequoia Capital says tax advisory is 80-90% intelligence work — rule-following, not judgment. Nate Jones argues most of what we call judgment is rules we've internalized so deeply we forgot they were rules. If they're right, the moat accountants keep defending on LinkedIn is real — but much narrower than the profession assumes.
The most valuable thing you'll do with AI this week is tell it it's wrong
AI matches experienced professionals 70-83% of the time. The profession's biggest complaint — that AI gets things wrong — is actually a description of where your professional value lives. The 17-30% where you catch the error, push back, and explain why? That's judgment. And it's the one skill nobody in the profession is developing deliberately.
Get it out of your head
The research was valuable. What happened next was more valuable — a 10-minute reflective conversation with the AI that turned a one-off project into a reusable methodology anyone in the firm can follow. That's context engineering. And it's how you stop losing what you learn.
Neither of us could have come to this position alone
A cross-border tax research project revealed a material legal error — not because AI got it wrong, but because the back-and-forth dialogue forced both sides to test every assumption. The best use of AI for complex professional work isn't asking questions. It's arguing until the answer survives.
Weekly AI Roundup: The AI is coming to you
Canopy is embedding AI bookkeeping directly into practice management — real-time GL visibility, issue detection, and month-end close automation inside the tool you already use. Plus OpenAI's $110 billion bet on agentic AI, Anthropic removing the last barrier to switching, and why 30,000 accounting firms watched one AI demo in three days.
90% cost savings. Productivity doubling. And your practice is still doing it by hand.
Marketing teams are seeing 90% cost reductions. Customer success teams are doubling productivity. And Anthropic's own data shows financial roles have an 85% theoretical AI capability with only 20% actual adoption. The gap between other industries and CAS practices isn't closing. It's widening — and it's measurable.
Your monthly close wasn't designed for one AI tool. It was designed for six.
Google Cloud says multi-agent orchestration is the architecture of 2026. Basis — now valued at over a billion dollars — is already running it for Top 25 accounting firms. The question isn't whether this model works. It's whether your practice will build it with your own specifications or let a vendor do it for you.