01The kid in the computer lab
I got hooked on computers in the late 1980s, in high school, right as personal computers were finding their way into ordinary classrooms. I wasn't just playing games on them — I was trying to make them do things, teaching myself to write programs in BASIC, then Pascal, then C. I assumed that was my future. When I headed to college, the plan was simple: major in IT.
Then I took an accounting class.
I've always been a numbers and math person, and something about accounting clicked in a way I didn't expect — the logic of it, the way everything has to tie out. The IT major never happened. I became an accountant instead.
But here's the thing about a detour like that: the tech interest doesn't go away. It just goes underground.
02The spreadsheet years
For the next few decades I was the accountant who couldn't stop building things. Macros and VBA in Excel — not because anyone asked, but because I couldn't stand watching a manual process eat hours that a few lines of code could give back. When companies I worked for implemented new systems, I was the finance person in the room who actually enjoyed it. I ended up doing multiple system implementations over my career, translating between what accounting needed and what the software could do.
Looking back, I was doing the same job the whole time: finding the gap between how the work was done and how it could be done, and closing it with whatever technology was available. For most of my career, "whatever technology was available" meant spreadsheets and ERP modules.
Then the technology changed.
03Early 2024: the ending that started everything
In early 2024 I was let go as Controller of a startup. That's the sentence version. The longer version is what anyone who's been through it knows — the sudden stop, the "what now."
What I did with the "what now" was go deep on AI.
ChatGPT had been out for over a year at that point, and like most accountants I'd poked at it. But with time on my hands, I went well past poking at it. My real testing centered on ChatGPT and Claude, run hard against actual accounting judgment calls, alongside a wider field of general AI tools that often weren't accounting-specific at all. And I was tracking the accounting-and-finance-specific side of the market closely too — I kept a running list, dozens of named tools deep, across bookkeeping, tax, FP&A, AP/AR, and close management. It was a gold rush: hundreds of AI applications launching every month, enormous claims everywhere, money pouring in. And when I ran the ones I actually put hands on through the only lens I have — would I trust this in a real accounting operation? — almost nothing survived.
I saw two things at once that spring:
The first was an opportunity. Businesses were going to need someone who could stand between them and the AI hype — someone who actually did the work AI was supposed to transform. I registered AI Business Concepts, started a YouTube channel, and began doing consulting accounting work to pay the bills while I tested whether the idea had legs.
The second was a problem: it was too early. Not slightly early — fundamentally early. There weren't enough tried-and-tested products that could be trusted with real accounting work. And accounting is famously slow to adopt new technology for a good reason: we're the people who get blamed when the numbers are wrong. Selling AI to accountants in 2024 meant selling something I couldn't yet fully stand behind.
I have proof of this, and I've left it up on purpose. One of my early YouTube videos — "Creating Excel financial models using Claude 3.5 Sonnet" — captures exactly where the state of the art stood. Claude could write VBA code for a full three-statement financial model, and it mostly got the formulas right. But watch what the workflow actually was: copy the generated code by hand, open Excel's VBA editor with Alt+F11, paste it into a module, run the macro, and hope. When a formula broke, you pasted the error back and waited for corrected code. That was the best tool available in 2024 — a clever engine with no car around it. And a demo on a blank workbook is a long way from something you'd trust inside a real month-end close.
I ended that video with a line that has aged well: "Although it is not perfect, it is a very good starting point that is likely to improve." Even I didn't appreciate how much.
And it wasn't a one-off. In a follow-up — "Creating Financial Models in Excel from a Visual using Claude" — I pushed further: I gave Claude a screenshot of a model layout I wanted and had it rebuild the model in Excel from the image (taking care to use a non-copyrighted layout — accountant's instincts). The labels all landed, most of the formulas worked, the colors didn't make it. The delivery mechanism was still hand-ferried VBA, but the pattern I use every day now was already visible: show the AI what you want, review what comes back, fix what's wrong. I even signed off asking viewers what else in finance and accounting they wanted covered with AI. Two years later, the mission hasn't changed — the tools finally caught up to it.
Most people's 2024 AI experiments are deleted. Mine are timestamped and public, and they're the single best evidence I have for the point of this whole story: the difference between then and now is not hype. I lived both ends of it, on camera.
After about six months, I made the honest call. The business idea was right; the timing wasn't. I went back in-house.
04Back inside — where the real education happened
In mid-2024 I took a Controller role at another startup. What draws me to startups is the building: standing up a finance department from the ground up, which is really a systems problem wearing an accounting costume. That role was where I moved from watching automation to running it — I started building Make automations into the department's workflows. One of them handled our weekly customer invoicing emails: for 20-plus customers every week, it drafted each email and attached the right invoice file automatically. The only human step left was review and send — which is exactly where the human step belongs.
That period taught me the lesson that now sits underneath everything I do: the value of automation only shows up when it's built from inside the workflow. The hundreds of AI tools I'd surveyed in 2024 were built from the outside in — impressive demos looking for a problem. What actually works is the reverse: start with the close checklist, the reconciliation, the reporting package, and pull the technology toward it.
Startups also come with uncertainty and stress — that's the trade, and I'd felt it twice now. In mid-2025 I chose stability and took a Controller role at a nonprofit, where I still work today. Different mission, same monthly rhythm: close the books, reconcile the accounts, report the results.
Which means that when the technology finally caught up, I was standing in exactly the right place to notice.
05The turn
For two years after that first too-early attempt, I kept checking back. Not occasionally — continually. I never wrote AI off, because the trajectory was obvious: it would keep getting better until it crossed the line from "interesting" to "trusted." The only question was when.
For me, the moment it crossed was the arrival of Claude's Cowork — an AI that doesn't just answer questions but works alongside you: reads your files, builds your spreadsheets, writes and runs code, and does it at a price point a solo practitioner or a lean finance team can actually justify. For the first time, I was looking at a complete application of AI that a general business audience could use — not a demo, not a point solution, not a six-figure implementation.
There's a symmetry here I didn't plan. The AI in my disappointing 2024 video was Claude 3.5 Sonnet — a chat window doing its best with a financial model. Two years later, it's Claude again that crossed the line for me — grown from that chat window into something that works the way an accountant actually works: in files, in spreadsheets, in repeatable processes. Same lineage, different era.
So I did what I've always done with new technology: I built something I actually needed.
My first real build was a daily task and meeting tracker — an application I now open every morning. One sentence: it keeps my projects and meetings organized, and lets me go back and review any decision that was made or ask questions about any point in a meeting — or across a whole series of meetings. I didn't write the code. I described what I wanted in plain English, reviewed what came back, and asked for changes until it was right. The hard part, it turns out, wasn't the technology. It was learning to describe precisely what I wanted — which is a skill accountants already have; we just usually aim it at staff instead of software.
Now I'm doing it again, bigger: building a month-end close application — checklist, reconciliation tracker, analysis, dashboard — and documenting the build as I go, including the dead ends. In 2024 I tried to make content about AI's promise. In 2026 I'm making content about AI's receipts.
06Why now, and where this goes
I'll be honest about what this is. AI Business Concepts is a side gig alongside a full-time Controller job — a way to share what I'm learning as I go, not a business plan with a countdown on it. I've enjoyed every part of the process, because this is the IT-kid-turned-accountant finally getting to do both jobs at once.
But I'm doing it publicly, and here's why. Two years ago I watched accountants correctly ignore AI because it wasn't ready. The danger now is that the same skepticism — which was right in 2024 — is wrong in 2026, and the people who hold onto it too long will be the last to benefit. I believe AI will be an integral part of accounting operations within two to three years. Not the fully-automated-accountant fantasy — the practical version: closes that take days less, reconciliations that surface exceptions instead of demanding them, reporting that regenerates itself, and accountants spending their hours on judgment instead of assembly.
Here's what makes this different: I'm not teaching AI from the outside. I run a real monthly close — reconciliations, accruals, the general-ledger mechanics, the exceptions that never fit the checklist — and I'm putting AI to work inside that, in real accounting work, every month. I've been waiting for this technology since the 1980s, tried it too early, told the truth about that, and now I write down what actually happens: what works, what breaks, and where a Controller's judgment still has to step in. Not a demo, not a framework — the receipts.
If you run accounting operations — inside a business or for clients — and you want to watch someone figure this out in real time, failures included: follow along.