One night shift: what an AI Employee does between 6pm and 7am

The easiest way to understand an AI Employee is to watch one work a shift nobody else is on. This is a night in the life of an Accounts Receivable Officer at a mid-sized business: what it does between six in the evening and seven in the morning, in what order, and what it leaves on the desk. The business and the figures are illustrative. The sequence is exactly how a role runs.

18:00. Everyone leaves. It reads.

The first thing a role does when the office empties is not work. It reads. Within the first hour it has been through every system its position description names: the ledger in Xero, the pipeline in HubSpot, the shared inbox, the SharePoint folders, the ERP, the bank feed, and the one supplier portal that has no API and has to be read through the screen the way a person would. It is not looking for a task yet. It is building the picture: who owes what, what came in today, which promises were made in email that the ledger does not know about.

This is the part people underestimate. A new starter takes weeks to know where everything is. An AI Employee reads everything the role touches, every night, before it starts.

19:30. The first task.

With the picture built, it looks at its objectives. This role's KPIs say debtor days under thirty-five and every invoice reconciled within twenty-four hours. Against those, the next task is obvious: the bank feed has ninety-six lines from today and eleven of them do not match a ledger entry. It matches the ones it can, with the reference and the amount, and writes the reasoning next to each one. Three it cannot match with confidence. Those go to the gate, not to the ledger.

21:00 to 03:00. The work most people never see done.

Every thirty minutes it asks what is next. It drafts the reminder for the invoice that went thirty days overdue today, in the tone the business uses with that customer, because it has read the last six months of correspondence with them. It notices a customer who paid a short amount against two invoices and works out which one they meant. It finds a credit note that was promised in an email in July and never raised, and prepares it. It reads the statement a supplier sent and checks it against the purchase orders, line by line, and flags the one that was billed twice. It updates the cash forecast for Friday with everything it now knows.

None of this was prompted. Nobody is awake. It is working to the position description, and every step is logged with the reasoning, so in the morning a person can see not only what was done but why.

What it did not do.

It did not send the reminder. It did not raise the credit note. It did not change a payment, a bank detail or a customer's terms, and it did not email anyone outside the business. Its position description says that anything over two thousand dollars, and anything that reaches a client, waits for a person. So the reminder is drafted and queued. The credit note is prepared and held. The three unmatched bank lines are sitting with a note that says what it thinks they are and why it is not sure. That is the gate, and it is a design choice, not a limitation.

07:00. The morning brief.

At seven the finance manager opens Teams and finds one message from the role. It lists what was done overnight, in eight lines, and then one thing that needs her: the credit note, with the July email attached, and three buttons. Approve, edit, decline. She reads it with her coffee, approves it, and the role raises it and moves on. The reminder goes out at nine because she approved that too. The three bank lines she resolves in two minutes because the hard part, working out what they probably were, is already done.

A person who did all of this would have started at seven and finished at lunch. Instead she started at seven with the decisions, which is the part of the job that actually needs her.