While 58% of professionals have tried AI, only 17% use it regularly. Kyle Ashcraft sees opportunity in that gap.
In episode 108 of the Earmark Podcast, host Blake Oliver sits down with Kyle, a CPA who built Maxwell CPA Review and helped over 1,500 students pass their exams, for a live demonstration that might change how you think about Excel automation. Their conversation shows how any accounting professional can start automating their work in under an hour. No coding experience required.
The AI Gap Nobody’s Talking About
“The more advanced AI becomes, we can take one of two directions,” Kyle explains during the demonstration. “You can continually veer away from it, and the more that comes out, you step farther and farther away from it. Or you can make it a goal to learn, let’s say, one new tool a week.”
The problem isn’t that accountants don’t want to use AI. It’s that they don’t have dependable strategies for implementing it. Kyle describes the typical approach as, “Opening up ChatGPT, throwing in a spreadsheet, and then giving it a prompt and seeing what it comes up with. Sometimes like a Hail Mary, where you just want to see if it gives you an acceptable output.”
There are two major issues with this approach. First, it often takes multiple attempts to get the output you want because ChatGPT can’t read your mind. Second, and this is crucial for accountants, when you upload a spreadsheet to ChatGPT, “your Excel document is going directly to OpenAI. Your prompt is going to them, and the prompt that they output to you is going to them as well.”
This matters because OpenAI’s data retention practices are questionable at best. They’re currently in a lawsuit with The New York Times and required to permanently retain logs. No wonder 70% of accounting professionals cite data security as their primary concern with AI adoption.
Enter “Vibe Coding”: When Everyone Becomes a Developer
Kyle’s journey started with a challenge. Could someone with zero coding experience build something that traditionally required a development team?
Four months later, he had his answer. Using Cursor, ChatGPT, and Claude, he built a complete assessment platform that identifies students’ weakest areas, emails follow-up practice materials, and provides analytics dashboards for professors. All with no programming background whatsoever.
“This really shows it’s possible to not have any idea what the code itself is saying, but with clear communication and patience, you can accomplish things that would have been impossible just two years ago,” Kyle tells Blake.
This phenomenon has a name: vibe coding. It’s coding without being a coder, using everyday language to generate complex scripts. During the demonstration, Kyle shows how Cursor generates hundreds of lines of Python code based on simple English instructions. You don’t need to understand what those lines mean, you just need to know what you want to accomplish.
Kyle offers a metaphor that reframes the entire relationship with AI. “Picture it like an orchestra and a conductor. You’re the conductor. You are in control. You set the tempo. You set the vision of what you want to achieve. And it’s the orchestra that’s doing all of the hard work.”
“There’s this assumption that AI is going to eliminate a lot of work,” Blake observes. “But what we find in reality is that it shifts the work from doing to reviewing. So that job is not going away, but now we review the output and provide feedback.”
The Script Solution: Privacy and Reliability in One Package
During a live Q&A, one attendee asks the question on everyone’s mind: “When you load the project into Cursor and it shows you the Excel files, does this AI platform not retain that client data? How is this different than uploading the Excel into ChatGPT?”
Kyle’s answer reveals why scripts are game-changing for accounting work. “It does not retain this data because with this process, it created this Python script, which is just Python code. It’s offline. There’s no record of this script.”
Your Excel data never leaves your computer. Instead, AI creates a script—basically a recipe—that runs locally on your machine. Think of it this way: instead of handing your sensitive client data directly to an AI company, you’re asking AI to write you instructions. The AI writes the instructions based on your request, but it never sees your actual data.
Blake highlights another advantage: “When Cursor communicates with AI services like Claude, it does so through APIs that have zero data retention policies. That’s in stark contrast to the chat interfaces most people use.” As he explains, these companies want large enterprises to be comfortable, so API interactions have much stricter privacy protections.
But privacy is only half the equation. Scripts also solve the reliability problem. Blake shares a cautionary tale about a Big Four firm in Australia that had to refund a government contract because its AI-invented citations didn’t exist. “They send an entire report to the government, the government clicks on the links for it, and they don’t exist. It’s disastrous if you don’t actually review the output.”
When another attendee asks about the risk of hallucinations, Kyle explains why scripts are different: “You’re not having an AI model interact with the Excel information. You’re having this step-by-step script that says, ‘do an auto sum of column B.” The script uses Excel’s own functions, it just automates the clicking and typing you’d normally do manually.
This deterministic nature means the same script produces the same result every time. As Blake notes, “We can reuse the script we created, apply it to a new Excel file and get the same expected result without having to check everything over again.”
The Three-Part Formula That Makes It All Work
“Goal. Steps. Output.” With these three words, Kyle unlocks the secret to making AI do exactly what you want.
During the demonstration, he tackles three real-world Excel challenges that every accountant faces. First up: a messy data export with empty rows, headers in row three, 14 different date formats, and inconsistent spacing.
His prompt is elegantly simple:
- Goal: Clean up this Excel file
- Steps: Identify any inconsistent formatting. Add basic color and style. Analyze each column to better understand its format
- Output: A new Excel document
Within moments, Cursor generates hundreds of lines of code. The result is a perfectly formatted table with consistent dates, proper headers, and professional styling. “It looks clean, smooth, with some nice shading,” Kyle observes. “It’s just easier to look at overall.”
When Blake asks whether Cursor can do its own checksum, they quickly add both files and ask Cursor to verify nothing was lost. The response: “All 20 transactions are present. All amounts were correctly processed. The sum of $19,000 is maintained.”
The second demonstration scales up the complexity. Kyle shows a General Ledger detail export with 400 rows spanning every account. Manually organizing this would require hours of filtering and copying. His structured prompt creates a summary tab showing account codes, transaction counts, debits, credits, and net amounts, plus individual tabs for each account’s detailed transactions.
“Instead of going to each account in your accounting system and exporting the GL individually, just export all the accounts together and then run this through,” Kyle suggests. What might take an hour completes in under a minute.
The third example addresses bank reconciliation, comparing statements to GL detail to find discrepancies. No more scrolling row by row. The automation identifies matching items, missing transactions, and differences between the files instantly.
Blake connects the dots for viewers. “I picture our listeners who work with some older ERP systems that don’t have very customizable reporting and who are doing a lot of manual formatting. Now you can automate that recurring task every month or every week.”
Getting Started Is Simpler Than You Think
The transformation begins with two downloads that take five minutes each. First, download Python, then download Cursor. Start with the free tier. Kyle uses the $20 monthly plan for daily use, but the free version is powerful enough to begin.
When you first open Cursor, it will ask you to install some packages like “pandas” for Excel interaction. Kyle recommends, “Click the dropdown button and choose ‘run everything’ so you trust the platform. It’s very reliable, and then anytime it needs a new required package, it automatically downloads that.”
Don’t forget to adjust your privacy settings. In Cursor’s settings menu, scroll to privacy options and select “privacy mode” with “no training data used.” This ensures your work isn’t incorporated into AI training datasets.
The key to success is to start small and be patient. “Try it with some information that is not private at all, maybe one of your own documents,” Kyle suggests. “The more patience I have, the more I follow up on that review step by giving it tiny pieces of feedback, the more it improves over time.”
Blake adds perspective on managing expectations: “When I try new tech, 80% of what I do doesn’t have a payoff, but then the 20% has a huge payoff. So don’t get discouraged if your first few attempts fail.”
For recurring tasks, the payoff compounds quickly. “Private roles always have month-end closing. Public end clients always need amortization and depreciation schedules for their notes,” Kyle notes. Even creating client checklists based on prior year information becomes a candidate for automation.
The Bottom Line: Your Move
The tools are accessible. The knowledge is available. As Kyle demonstrated with live examples, you can go from messy data to polished reports in minutes using nothing more than clear English instructions.
So, will you step away from AI as it advances, or learn one new tool at a time and stay connected to this movement? Because as Kyle reminds us, “It’s not going to go away. It’s just going to become more integrated into everyday work culture.”
To hear these demonstrations in action, listen to the full episode at podcast.earmarkcpe.com/108. Kyle has also offered to help early adopters, so reach out to him at kyle@maxwellcpa.com with questions or to brainstorm how this could apply to your specific work situation.
As Kyle challenges at the session’s close, “Try your first task with it this week and see how it works for you.” The revolution in accounting work is here, waiting for those bold enough to embrace it.