It reads before it acts
An assistant that acts on a room full of customers has to earn the right to act. Ours earns it the same way a good manager would: it looks first, says what it is about to do, and waits.
Put happy hour on the bar TVs from three to six on weekdays, then sports after seven. That is the whole instruction. What happens next is the part that matters.
It reads
The agent looks at the screens before it says anything. Ask what the bar is showing and it goes and looks, and every answer names the screens it found: Bar left, Bar centre and Bar right are online and on the zone programme. It does not answer from memory. It does not describe a wall it has not seen.
It proposes
A change arrives as a card that says exactly what will happen. The programme it would write, window by window. The screens it would touch. Two buttons: Approve, Not now. Nothing has moved yet. If the card is wrong, nothing was wrong on the wall, because the wall was never involved.
It does the whole job
This is not a search box with a personality. It programmes a zone, builds a playlist, puts the weather on the patio, starts trivia at eight, follows a team, spreads today's games across every screen and keeps the bar centre on specials. It reaches the same tools your staff do from the app, which is why it can finish a job rather than describe one.
It knows your venue
Your screens, your zones, your menu, by name. It never asks you what a TV is called, because it already knows, and it never asks which loop is on the patio, because it can see. When you say bar center it finds Bar Centre. When you name a screen that does not exist, it tells you the ones that do rather than guessing.
It still asks first
You will hear a lot about agents that act autonomously. In a room with customers in it, that is not a feature. The right amount of autonomy for a wall of televisions is: do all the reading, all the working out, all the building, and then show a person the one card that changes what the room sees. That is where we put the line, and it is why staff actually use it.
← All notes