In more than 500 Chipotle restaurants, the crew now gets a daily instruction from software. It works out how busy the restaurant will be, then tells people when to grill the chicken, simmer the beans and chop the produce. It takes over a guess that a manager used to make, and the sheet where someone typed that guess in. Chipotle plans to put it in every restaurant in early 2027.
A forecast is the safest job to give an AI
Four things make this decision easy for an AI. It repeats many times a day, so there is a lot of practice. The answer is checked by evening, because the steak ran out or too much rice went in the bin. The history it needs is already in the till. And the person who sees a bad forecast is an employee who can overrule it, so the cost of a mistake is some wasted rice or a short wait for more steak.
Voice ordering has none of that protection. When McDonald's tried again after its 2024 shutdown, it chose a new system rather than giving up on the idea, but the problem is the same: the customer is the one who hears the error. That is why Square's talking kiosk shows customers what the AI heard. When the customer is the first to see a mistake, the design has to let them fix it on the spot.
The same pattern shows up at Starbucks, and next at the manager's desk
Starbucks is putting AI into its supply chain to make ordering and restocking more accurate, and in 2027 it will test AI that listens in the background of drive-thru orders to help baristas get them right. The AI helps the person who takes the order. It does not take the order itself.
Chipotle is moving the same way. Ava Cado began in 2024 as a chatbot for job applicants, and Chipotle is now building it into an assistant that gives restaurant managers advice during the shift. If it works like the prep tool, managers will get suggestions from software and decide which ones to follow.
That changes what a good employee is. A good cook used to be partly valuable for guessing the lunch rush right. When software makes that guess, the skill that matters is noticing when it is wrong: a school event down the road, a storm that keeps people home, a delivery that arrived late.
What this looks like in a bakery
Take a neighbourhood bakery that throws away 20 loaves a day at $3 each. That is $60 a day, about $1,800 a month, and it still sells out by ten o'clock on busy mornings. The owner exports two years of till records, grouped by weekday, with rainy days and holidays marked. She gives them to an AI assistant and asks how many of each item to bake on each day of the coming week.
For two weeks she compares the forecast with what she would have baked, and writes down what was left over and what sold out. A shop cannot track ingredients live the way Chipotle's kitchens do, but a forecast by weekday and weather needs only records she already has. If it halves the loaves in the bin, that is $900 a month back, plus fewer empty shelves at ten. She still decides how much to bake. The difference is that she is no longer deciding from memory.
The first place AI gets trusted at work is the decision that repeats every day and can be checked by closing time. Most businesses have one, and finding yours is the best place to start.