Seven tasks an AI agent closes in a restaurant today
An agent closes tasks that have an input, a rule and a checkable result: receiving documents, matching them to the catalogue, arithmetic checks, price control, dish cost, preparing the reconciliation and answering questions from the data. Decisions about suppliers, menu prices and people stay with the team - the agent prepares them, it does not take them.
The word «agent» usually means a showcase. It is more useful to look at the list of tasks: what it actually does on its own, and how you know a task is closed.
Seven tasks
Receiving the document. A supplier email, a photograph of a delivery note, an e-invoice, PEPPOL. The document enters the workflow within the hour, without being forwarded to the bookkeeper in a chat.
Recognising the lines. Not just the document total but every line: item, quantity, unit, price, discount, VAT.
Matching to the catalogue. The supplier's line is tied to your item, with the pack and the multiplier: a case of 12 becomes kilograms in stock.
The arithmetic check. Quantity, multiplier and price must agree with the line total, and the lines with the document total. A document that does not add up waits for a person and never reaches the books.
Price control. The price is converted to a base unit and compared with your history for that supplier - see the purchasing signals page.
Recalculating cost. A new price pulls through to every dish that contains the item and shows which one went past target.
Preparing the reconciliation. Differences with the accountant's statement are sorted into baskets as documents arrive, not on the first of the month.
Where the line runs
The agent does not decide whether to change supplier, whether to raise a menu price, or what to do about an employee. It brings the grounds: the line, the amount, the date and the history. The decision is human, and so is the responsibility for it.
The second line is writing. A document reaches the accounting system only after a person has confirmed it. This is not caution for its own sake: a wrong quantity travels quietly into stock, where nobody checks it against the paper.
How you know a task is genuinely closed
The task has a counter: how many documents went through on their own, how many needed a person.
Every exception carries an amount in euros - otherwise it cannot be prioritised.
An error shows up as an error, not as a tidy average.
People handle exceptions, not the flow.
The connection sequence and the roles in the team are on the how it works page.
How it worksPurchasing signalsHow is an agent different from ordinary automation?
Automation runs a fixed script. An agent chooses which data it needs for a check and surfaces only the exceptions.
What happens to a document that does not add up?
It stops and waits for a person. It does not reach the books.
How many documents go through without a person?
It depends on how well the catalogue is matched. The pass-through counter is visible in the app, per supplier.