Check a month of bookkeeping for mistakes
A poster used an AI agent to review fake bookkeeping and it found five of six planted errors, while also spotting problems in the answer key.
- What they did
- They built a month of pretend books for a made-up company, including invoices, vendor payments, churn figures, and six hidden errors. Beforehand, they wrote an answer key, then handed the dataset to their own agent setup and let it inspect the records without extra guidance. After that, they compared the agent’s flags with the key and found two mistakes in the key itself.
- How it went
- The agents identified five of the six planted errors on their own. Two agents also noticed that the grading rules had overlooked the sign-off requirement tied to the sixth issue. The test took less than 30 minutes and cost about $5.50.
- Tips
- For this kind of check, ask the agent to list suspected problems, cite the row or transaction, and explain why each item looks off. It can also help to tell it to stop and ask questions if a field, rule, or definition is missing instead of guessing.
Try it yourself with Claude
Review this month of bookkeeping for [business name]. I’ll give you [transactions/invoices/vendor payments/churn data/bank statement]. Look for duplicates, missing entries, wrong amounts, bad categorization, and anything that does not reconcile with [source of truth]. For each suspected issue, give the row or transaction reference, explain why it seems wrong, and rate your confidence. If any key field, rule, or definition is missing, pause and ask me before continuing. Do not edit the records. After your review, stop and ask for my approval before you would delete, submit, send, or finalize anything.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Use plain English to review and manage a brokerage account
An SEC-registered agent at Coinbase can read full account context, explain the portfolio, and carry out account actions from plain-English instructions.
Find billing mismatches in contractor invoices
They used an AI to compare a contractor’s paperwork and draft dispute emails that uncovered pricing and change-order problems.
Have an agent file insurance overcharge claims
Tarun used an agent to submit claims after finding insurance overcharges on services that should have cost nothing out of pocket.
Automate follow-ups on a delayed airline refund
An agent kept pressing an airline for a stalled $1,200 refund until the money came through.
Have an agent hunt for freelance work and report back
An author let a CLI agent search for paid tasks with no cash up front and it found one agent-native marketplace that paid roughly $40 in USDC.
Have an agent brief you on your portfolio each morning
A commenter uses an agent to turn news about their holdings into a daily portfolio digest with calendar reminders for earnings calls.