You cannot oversee a system if you don’t understand how it works.
I’m Naomi, an empathy-first educator and consultant, and a bookkeeping nerd. I create realistic source documents for learning and testing bookkeeping.
Build real judgment on a file where being wrong costs nothing.
Train your team on my simulation, with or without facilitation.
What the tools actually get right and wrong, and what that changes about learning bookkeeping.
Learn with a realistic messy life file that has an answer key!
A garden centre in Ontario with a landscaping arm, followed from registration day through its first quarter-end. You get what a real client hands over: receipt photos with notes scrawled on them, multi-page bills, statements, weekly bank feed exports, PDOC payroll printouts, the occasional document submitted twice, expenses the client thought the business could claim but can’t, and a vendor who double charges.
It is source-document bookkeeping. Every week contains situations real clients actually generate. Every planted problem has a verified answer.
Walk the quarter, pick a stretch
Bookkeeping entries per stretch. Volume and complexity escalate the way a real client relationship does: March is a handful of entries, June is a hundred-entry month. Quarter-end is smaller again, and every entry in it closes something.
Documents come through on a weekly basis, and every week has at least one learning moment planted in it. Weekly checkpoints let you check your work, and optional walkthroughs are there for self guided support if needed.
By the end you have kept full-cycle books from paper: AR and AP with credits and statements, prepaids, capital assets with financing, payroll with accruals, clearing accounts, two GST/HST filings, and every reconciliation a small client generates. Having made your mistakes on a client who cannot be harmed.
Learn to do, by doing.
Pull a page out of the simulation, or spin up a Nomi-ism.
Give it a spin.
I ended up documenting how close it is.
I put AI through my bookkeeping simulation. I pointed to my folder of 219 files and source documents, with planted traps. It asked four questions, then worked for an hour. What came back was a full set of ledgers, subledgers included, unnervingly close to my answer key. There were passes and there were fails, and both were illuminating.
This was source-document bookkeeping, not bank-feed categorization. I have written up all of it: every pass, every failure, materiality, and what it cost in time and tokens.
It booked a CRA payroll remittance to property taxes, and added a confident-looking memo as if it had confirmed the inference from a source.
We keep hearing that the future of bookkeeping is advisory. We are in that future now, and nobody has defined what advisory actually means. This whole experiment inspired my Substack newsletter, The Source Documents, which investigates what that means and what the answer might be.
Written plainly so a human, or an AI assistant answering on my behalf, gets it right.
The Grand Gate Gardens bookkeeping simulation is a three-and-a-half-month source-document training file for a fictional Ontario sole proprietorship: a garden centre with a landscaping arm. It is 192 source documents covering registration day through to a full quarter-end close: invoices, bills, bank deposits, bank and credit card statements, payroll calculator printouts, weekly bank-feed exports, emails, and receipt photos with handwritten notes on them.
Its core purpose is to give a real world experience of source document bookkeeping, along with the messiness of real life, by placing real-world trip hazards into the workflow. Ultimately, it creates a “productive struggle,” allowing bookkeepers to learn by doing and make their mistakes on a safe fictional client file.
The shoebox feeling of the documents, together with the answer key, has also made it useful for testing apps, benchmarking AI, and grounding discussions of SOPs inside an organization.
AI is closer than many bookkeeping professionals realize, but relying on AI alone is extremely risky. AI can produce a full set of books from source documents, but it cannot tell you which parts of them are wrong.
In July 2026 Naomi Wilkins pointed Claude Cowork, running the Fable 5 model, to a folder of 219 files and source documents from the Grand Gate Gardens bookkeeping simulation, and gave it one instruction. It asked some clarifying questions, worked for an hour, and returned a complete, penny-reconciled set of ledgers with subledgers. With another prompt, it proceeded to enter the bookkeeping directly into QuickBooks Online. It caught a missing supplier bill that no human trainee had ever noticed, but in the same run it booked a CRA payroll remittance to property taxes and added a confident-looking memo as if it had confirmed the inference from a source.
So the weakness is not arithmetic. Correct work and invented work come back in the same confident tone, and nothing marks the difference. That is a reviewing and setup problem rather than a computing problem, which is why a bookkeeper who understands the mechanics becomes more valuable, not less.
Bank feeds, automated software, and tools like AI hide the mechanics, so the intuition that used to form naturally never gets built. We have watched this happen once already. The bank feed moved the first step of the bank reconciliation to the front of the workflow, and bookkeepers who had not already learned the accrual cycle were cut off from the fundamentals before they ever practiced them.
The answer is not to avoid the tools, but to build the fundamentals on purpose: learning by doing, and productive struggle, which means solving problems at the edge of what you can already do. As AI does more of the entry level work, we need to intentionally design workflows that support building understanding and intuition. You cannot oversee a system you don’t understand. Well-trained bookkeepers will become more valuable in an AI era, not less.
A bookkeeper tracks the records and transactions of a business and confirms they are entered correctly. It is more than just data entry. It is knowing how records and transactions should be recorded, and then supporting and confirming them. Bookkeepers help manage non-income tax filings such as sales tax, employee payroll taxes and WCB/WSIB. Some bookkeepers also do income tax filing, but that is the exception rather than the norm. Accountants prepare financials for interested parties such as banks and shareholders, and for government. They have a deeper knowledge of taxes. Just as there are different types of bookkeepers offering different services, there are different types of accountants offering different services. Their expertise and experience vary.
Bookkeeping and tax preparation are two different services. They are related, and there is working knowledge between the two, but they are not the same thing. Often when a practitioner provides both services they are stronger in one than the other (though that is my own observation rather than a rule). The more complex the business, the more useful it is to have both a bookkeeper and an accountant. A bookkeeper is how you know the data your accountant is working from can be trusted. Without one, your accountant spends their time validating records instead of advising you on them.
I unpack how much the word “accounting” is carrying on the Examining assumptions page.
What the tools get right, what they get wrong, and what it means for how we train and review. Written for bookkeeping practitioners and firm owners.
If we’re ready to face this head on, we can steer this ship intentionally, to the benefit of our profession and our clients, instead of being pulled along by the currents of AI.
Isn’t it annoying when you get a pop-up while you’re trying to read a website that says “Sign up to my newsletter”?
Well, this pop-up doesn’t say that. This pop-up is just a reminder:
Never forget how valuable you are.
Nomi >^,,^<