LO
Waiter / restaurant server — tasks, one by one
The unit of analysis is the task, not the job title. Each one below carries its direction, whether the judgement rests on evidence or on platform inference, the reasoning, and what it does not establish.
Every task on this page#
Taking the order
Automating✓ Evidence-backedRecording what each person wants, with the modifications, the allergy, and the one they changed their mind about.
QR-code ordering moved a large share of this to the customer's own phone during the pandemic and it stayed, which makes it the clearest completed automation in this job — and it was not done by AI. The saving is real because the task is pure data entry once the menu is digital.
Where this has happened, the covers per server went up and the job got harder rather than smaller — the server now handles more tables and has lost the thirty seconds at the table where they used to read the room. Removing a task can degrade the job it was part of, and this is the cleanest example of it in the service group.
Running the room
Still human-led✓ Evidence-backedHolding twelve tables in your head at once: who is waiting, who needs to be left alone, which course goes out next and which table will complain in five minutes if nobody goes over.
The scheduling part of this is genuinely computable and some systems attempt it; the input is not. Knowing that table six has gone quiet in a bad way rather than a good way is a reading of people that nothing in a restaurant is instrumented to capture, and the decision it drives — go over now, or leave them — is made in seconds with no data.
This being human does not set how many tables one person is given. The same judgement performed across eighteen tables instead of ten is a different job at the same job title, and that ratio is set by a manager after the ordering tablet arrives — which is how automation reaches this task without touching it.
Carrying things through a crowded room
New task✓ Evidence-backedPlates, trays, drinks — through a space full of moving people, chair legs and children.
Running trays is the one task in this job that robots do at scale today: delivery robots are deployed in thousands of restaurants, mostly in Asia, and they work because the route is fixed and the failure mode is stopping. What they do not do is the last metre — putting the right plate in front of the right person and clearing it.
A running robot in a dining room is as often a marketing decision as a labour one, and where it has been deployed the reported effect is usually on how far staff walk rather than on how many staff there are. Read this direction as the technology working, not as a headcount result — what has been measured is the walking, and no operator has published a service roster before and after.
When it goes wrong
Still human-led≈ Platform inferenceThe wrong dish, the hour-late main, the allergy that was on the ticket and not on the plate — and the decision about what to take off the bill.
Recovery is the whole of hospitality's reputation and it is performed in person under time pressure, with the authority to give something away. That authority is the mechanism: the customer is being shown that a person with power over the bill is sorry, and a screen cannot perform that even if it can issue the refund.
The authority is the thing, and it is withdrawn more often than it is automated: chains that centralise comp decisions into an app remove this task from the server without a robot being involved at all. What looks like a durable human task can be hollowed out by a policy change in an afternoon.